This appeal arises from a long-standing dispute between the parents of Gregory K. and the Longview School District as to what constitutes “free appropriate public education” for Gregory under the Education for All Handicapped Children Act, 20 U.S.C. § 1400 et. seq. Under the Act, the states receive federal funds to provide special education services. Procedures and standards for school districts to use in providing the special education services required under the Act are established by the states.

BACKGROUND

Born in 1968, Gregory K. entered kindergarten in the Longview public schools in the fall of 1973. He repeated kindergarten during the school year 1974-75. In October 1974, Gregory was referred to the District’s Special Education Department for evaluation, including an IQ test on which he scored 85. At that time, students were eligible for special education only with IQ scores below 75, so Gregory remained in a regular classroom.

In January 1976, while Gregory was in first grade, the District again assessed him for special education. He was found eligible, scoring 70 on an IQ test. The District developed an “individual education plan” for Gregory that recommended placing him in an “educable handicapped classroom,” a program for mildly retarded students. Gregory’s parents consented to placing him in this class. In June 1976, the District reevaluated his special education placement. Based on an IQ test score of 69 and his school performance, the District recommended that Gregory continue in the special education class. According to his parents, however, Gregory was unhappy and not learning in the class, so they chose to have him repeat first grade in a regular class during the 1976-77 school year.

During 1977-78 and 1978-79, Gregory was in regular second and third grade classes. He had great difficulties, and in May 1979 the District again assessed Gregory. His IQ was 69 and his academic performance was less than half the expected grade level. The parents had Gregory tested by Dr. Loveland, a clinical psychologist. The results of his testing were consistent with those of the District’s testing, but he advised the parents against placing Gregory in a special education class for mildly retarded children. The parents again refused the District’s placement. For the next two school years, 1979-80 and 1980-81, the parents placed Gregory in regular classrooms in a private school.

In the summer of 1981, Lois Lewis, an employee of the District’s Indian Education Program, began tutoring Gregory. The parents believed that under her tutoring Gregory’s learning progress increased. When they re-enrolled Gregory in the District in the fall of 1981, they asked that the tutoring continue. The principal of Gregory’s school agreed, but advised the parents that the District could offer a special education program more appropriate for Gregory than the tutoring alone. In the summer of 1982, Dr. Levinson, a neurologist, examined Gregory. Although he did not use an IQ test, he concluded that Gregory was dyslexic, not mildly retarded.

The Indian Education Program in which Lois Lewis worked as a tutor was reorganized in 1982-83, and she resigned. In February 1983, the District told the parents that a new tutor would begin working with Gregory. The parents alleged that this change constituted a change of placement in violation of the Act’s “status quo” provi sion. They alleged that the proposed new tutor was not qualified as a special education teacher and that he would not be giving Gregory as much individual tutoring as he had received from Mrs. Lewis. The parents requested a hearing and removed Gregory from school for three hours each day to be tutored privately by Lois Lewis. They paid for this tutoring themselves.

The District again offered to assess Gregory for special education. After a dispute over testing was resolved, Gregory’s IQ was found to be 66 and his academic performance in all areas was less than three-fourths the expected grade level. The parents disputed the District’s conclusion that these results satisfied the state’s criteria for “mildly mentally retarded.” They believed that Gregory had not been properly tested for learning disabilities. The parents rejected the District’s “individual education program” for Gregory based on this assessment.

In August of 1983, a hearing was held on this dispute. The parents and the District agreed to an interim schedule for Gregory. The schedule followed the District’s 1983 IEP and included a mix of regular and special education classes, plus an hour of private tutoring each school day. The parents refused to accept any District tutors, however, and again removed Gregory from school for three hours of tutoring by Lois Lewis every day.

On December 7,1983, the hearing officer issued findings of fact and conclusions of law, affirming all the District's actions. He ruled that an appropriate schedule for Gregory was four periods of regular education classes and two hours of special education classes per day. The state Superintendent of Public Instruction upheld these rulings.

The parents appealed to federal district court. After a bench trial on August 23, 1985, the court ruled: (1) that Gregory K. was not “mildly mentally retarded,” but had a learning disability; (2) that the District’s proposed placement was inappropriate; (3) that the tutoring arrangement begun in 1981 was “special education” under an appropriate “individual education program”; and (4) that the District violated the Act by improperly changing the tutoring program in 1983. The trial court ordered the District to grant Gregory academic credit for the tutoring and to reimburse Gregory’s parents for the tutoring they had paid for themselves. The court denied the District’s post-trial motions to dismiss damage claims and to require a three-year reassessment of Gregory.

STANDARD OF REVIEW

The Court of Appeals reviews the district court’s findings of fact for clear error. Burlington Northern, Inc. v. Weyerhaeuser Co., 719 F.2d 304, 307 (9th Cir.1983). “A finding of fact is deemed clearly erroneous when although there is evidence to support it, the reviewing court is left with a definite and firm conviction that a mistake has been made.” Id. Even where, as here, the district court relies on a written record of administrative proceedings, the clearly erroneous standard applies to findings of fact. Id.

The Court of Appeals reviews de novo the district court’s conclusions of law. Id. Unless a mixed question of fact and law is primarily factual, we review mixed questions de novo. United States v. McConney, 728 F.2d 1195, 1199-1204 (9th Cir.) (en banc), cert. denied, 469 U.S. 824 , 105 S.Ct. 101 , 83 L.Ed.2d 46 (1984). For example, whether the school district’s proposed IEP was a “free appropriate public education” as required by the Education for All Handicapped Children Act is a mixed question that we review de novo. Wilson v. Marana Unified School District, 735 F.2d 1178 , 1181 (9th Cir.1984); Department of Education, State of Hawaii v. Katherine D., 727 F.2d 809 , 814 n. 2 (9th Cir.1983), cert. denied, 471 U.S. 1117 , 105 S.Ct. 2360 , 86 L.Ed.2d 260 (1985).

III

DEFERENCE TO STATE ADMINISTRATIVE FINDINGS

The District contends that the trial court failed to give proper weight to the state administrative findings in this dispute. Because the key issues are matters that we review de novo, we need not consider how much weight the trial court gave or ought to have given to the administrative findings. Instead, we address this issue to determine how much weight we are to give the administrative findings on those matters that we review de novo.

As the Supreme Court has held, courts must give “due weight” to judgments of education policy when they review state hearings under 20 U.S.C. § 1415 (e). Hendrick Hudson Dist. Bd. of Ed. v. Rowley, 458 U.S. 176, 206 , 102 S.Ct. 3034, 3050 , 73 L.Ed.2d 690 (1982). Citing Rowley , we have held that “courts should not substitute their own notions of sound educational policy for those of the school authorities which they review.” Wilson v. Marana Unified School Dist., 735 F.2d 1178 , 1183 (9th Cir.1984). How much deference to give state educational agencies, however, is a matter for the discretion of the courts:

Town of Burlington v. Dept. of Ed., 736 F.2d 773 , 792 (1st Cir.1984), aff'd, 471 U.S. 359 , 105 S.Ct. 1996 , 85 L.Ed.2d 385 (1985).

ELIGIBILITY CRITERIA

The District challenges the trial court’s ruling that “Gregory K. has a learning disability” and “is not mentally retarded.” This is a mixed question of law and fact. The undisputed facts are the results of much testing Gregory received over the years from the District and from experts consulted by his parents. We apply to these facts the state regulations for determining eligibility for special education under the Act.

Students are eligible for special education under the Act if they have any of several handicaps as defined by state regulations. Washington regulations define a “specific learning disability” as:

WAC 392-171-406 (1983) (emphasis in original).

Turning to the “eligibility criteria,” we find a threshold requirement of “a standardized individual test designed to measure intellectual functioning,” an IQ test of the sort Gregory was given many times. WAC 392-171-411. This is the same threshold requirement for determining mental retardation. See WAC 392-171-421(l)(a). What distinguishes the eligibility criteria for a learning disability from those for mental retardation is that the “measured level of intellectual functioning” of a student suspected of having a learning disability “must be near normal or above.” WAC 392-171-411. Thus, Dr. Levinson’s diagnosis of dyslexia, even if accurate, is not by itself enough to meet the criteria for a learning disability. The “eligibility criteria” for mild mental retardation are as follows:

WAC 392-171-421(2)(a) (1983).

The record provides no support for the trial court’s conclusion that Gregory had a learning disability, not mild retardation, as defined in the above regulations. None of Gregory’s IQ scores is in the “near average, average, or above average” range. Only the first of Gregory’s IQ tests resulted in a score above 75. That score was deemed less reliable by school psychologists because Gregory was only six at the time. In Gregory’s subsequent testing, the resulting IQ scores are consistent: IQ of 70 in 1975; 69 in 1976; 69 in 1979; 69 in 1982; and 66 in 1983. All of these scores are within the mildly retarded range.

The 1979 testing by Dr. Loveland, the parents' witness, is consistent with the District’s evidence as to Gregory’s intellectual functioning. Dr. Loveland’s report at the time stated that “the total results of the evaluation are indicative of a generally low potential across most areas assessed rather than it being indicative of any serious or specific learning disability as such.” Dr. Loveland’s subjective conclusion was that Gregory would not benefit from placement in a special education class. But the state regulations require classification on the basis of objective data.

In addition to an IQ in the 51-75 range, the eligibility criteria for the mildly retarded classification include deficient academic and adaptive behavior. Thé District offered evidence that Gregory was at or below three-fourths of grade level in all areas. The parents did not challenge the District’s evidence as to Gregory’s academic performance. The 1983 testing by Dr. Loveland, their witness, showed academic performance in all areas at or below three-fourths of grade level. The parents also did not dispute that Gregory’s adaptive behavior met the criteria for mild retardation.

The trial court offers no factual or legal explanation for its ruling that Gregory K. is learning disabled rather than mildly mentally retarded. We are “left with the definite and firm conviction” that the trial court erred in its weighing of the evidence regarding Gregory’s IQ testing. The only test result above 75 was an early, less reliable test. Dr. Levinson, the expert witness who testified that Gregory was dyslexic and not mildly retarded, did not give Gregory an IQ test. The objective testing done by the family’s other expert witness, Dr. Loveland, corroborated the District’s testing results. These facts fail to support the threshold eligibility criterion for a ruling that Gregory is learning disabled. Applying the state’s regulations to these facts as we must, we hold that Gregory’s eligibility for special education was correctly based on mild mental retardation.

THE EXISTENCE OF AN INDIVIDUAL EDUCATION PROGRAM

Gregory K. was eligible for special education under the Act. The District does not dispute this. He was first found eligible in 1976 and will remain eligible until he graduates from high school, turns 21, or no longer needs special education. See WAC 392-171-331(1). The parents’ rejection of certain special education classes does not determine whether Lois Lewis’s tutoring was “special education” as defined by the Act. On the other hand, Gregory’s eligibility for special education under the Act does not mean that whatever services he received from the District were automatically “special education” entitled to protection under the Act.

The key question is whether Gregory’s education in the District at the time of this dispute was being provided under “an individualized education program.” “Free appropriate public education” as mandated by the Act is properly defined by state regulation to mean “special education and required services which ... [a]re provided in conformity with an individualized education program which meet [sic] the requirements of WAC 392-171-461.” WAC 392-171-310(l)(c). The trial court ruled that such an IEP “was in place for Gregory K. from September 1981, through February 1983.” This was the period during which Gregory was enrolled in the District, attending regular non-academic classes and being tutored in all academic subjects by Lois Lewis. The existence of an IEP was necessary to the trial court’s further determinations that this program was an “appropriate” placement for Gregory, that the District acted improperly in changing this placement in February 1983, and that the parents were entitled to reimbursement for their expenses in maintaining the placement.

Although denominated a finding of fact, the court’s ruling on this issue involves a mixed question of law and fact. The law we must apply to the facts of the ease is the regulatory prescription for an IEP:

WAC 392-171-461.

The trial court gave no factual or legal basis for its ruling that an IEP was in place from September 1981 through February 1983. The parents argue that several documents in the record contained the functional elements of an IEP. One of these is a letter from the school counselor to the parents acknowledging their request that Lois Lewis continue tutoring Gregory. The other is a memo from Lewis to the parents summarizing her goals for Gregory. Lewis’s memo cited no “assessment analysis” as the basis for her work with Gregory; none of the District’s special education personnel familiar with past assessments of Gregory were involved in preparing the memo. Moreover, there was no statement of the extent to which Gregory could participate in regular classes, or what those regular classes should be. There was no statement as to the anticipated duration of the tutoring. There was no statement as to the objective criteria and procedures for determining whether Lewis’s goals for Gregory were being met. In short, the memo upon which the parents relied failed to meet the comprehensive requirements of an IEP.

The parents also argue that even if these documents are inadequate for an IEP, the District should be estopped from denying the existence of an IEP because it permitted the parents to believe that Gregory was receiving special education. The parents cite some “IEP” forms from the fall of 1982, but those were only for speech therapy, not for a full assessment. There is no evidence that the District believed or led the parents to believe that Lewis’s tutoring was a special education placement pursuant to an IEP.

At oral argument, the parents contended that an IEP from January 1976 applied to the tutoring Gregory was receiving from Lois Lewis. We understand but reject this contention. That IEP prescribed no tutoring such as Lois Lewis provided. Instead it recommended placing Gregory in an “educable handicapped classroom.” Gregory was placed in such a class for awhile, but his parents eventually asked that he be returned to a regular classroom. The 1976 IEP establishes that Gregory was — and still is — eligible for special education. It does not, however, provide protection under the Act for the tutoring Gregory received from Lois Lewis.

The program of non-academic classes and academic tutoring that Gregory received from September 1981 through February 1983 was not under an “individualized education program” as prescribed by the Act and by state regulation. Because Gregory’s tutoring program was not special education pursuant to an IEP, we need not decide whether the District’s actions in February 1983 constituted a change of placement. The “status quo” provisions of the Act apply only to special education received pursuant to an IEP. See Zvi D. v. Ambach, 694 F.2d 904, 906 (2d Cir.1982).

AN APPROPRIATE PLACEMENT

Although we have held that the District did not improperly change Gregory’s placement in February 1983, we must also decide whether the placement proposed by the District later that year was “appropriate.” The trial court ruled that the placement was not appropriate. The trial court gave no factual or legal basis for its ruling on this issue. This issue more than most others in the case involves educational policy. The “due weight” that courts must give to the state education agency’s findings on matters of educational policy should prompt courts in the future to provide a more thorough explanation when reversing an agency’s ruling on the appropriateness of a special education placement.

We review de novo the appropriateness of a special education placement. Wilson v. Marana Unified School District, 735 F.2d 1178 , 1181 (9th Cir.1984); Department of Education, State of Hawaii v. Katherine D., 727 F.2d 809 , 814 n. 2 (9th Cir.1983), cert. denied, 471 U.S. 1117 , 105 S.Ct. 2360 , 86 L.Ed.2d 260 (1985). An “appropriate” public education does not mean the absolutely best or “potential-maximizing” education for the individual child. See Hendrick Hudson Dist. Bd. of Ed. v. Rowley, 458 U.S. 176 , 197 n. 21, 200, 102 S.Ct. 3034 , 3046 n. 21, 3047, 73 L.Ed.2d 690 (1982). The states are obliged to provide “a basic floor of opportunity” through a program “individually designed to provide educational benefit to the handicapped child.” Id. at 201 , 102 S.Ct. at 3048 . Gregory’s parents believe emphatically that Lois Lewis’s tutoring has helped him learn. Our de novo review, however, must focus primarily on the District’s proposed placement, not on the alternative that the family preferred. Even if the tutoring were better for Gregory than the District’s proposed placement, that would not necessarily mean that the placement was inappropriate. We must uphold the appropriateness of the District’s placement if it was reasonably calculated to provide Gregory with educational benefits.

We have already held that the District correctly assessed Gregory as mildly mentally retarded, as that handicap is defined by state regulation. We therefore must reject the parents’ blanket contention that any placement based on that assessment had to be inappropriate. Evidence in the record demonstrates that the District’s pro posed placement was designed to match the strengths and weaknesses that emerged from testing of Gregory’s achievement. The District’s April 1983 assessment recommended that Gregory be placed in regular non-academic classes, such as physical education, and in special education classes where he could work at his achievement levels in various academic subjects. Because his math computation abilities were near average, the report stated that a “mainstream basic math course” might be more appropriate than a special education program in math. The IEP based on this assessment set forth specific, realistic goals in all subjects based on his varied achievement levels.

The District’s IEP reflects the Act’s preference for the “least restrictive” program for handicapped students. See WAC 392-171-471. The District’s assessment and IEP, corroborated by Dr. Loveland’s 1983 testing, also support the District’s longstanding concern that in Lewis’s individual tutoring, which emphasized rote learning, the more mechanical aspects of Gregory’s language and math skills were developing. But other, more complex intellectual skills were not developing. His ability to recognize and spell individual words, for example, was considerably greater than his general reading comprehension and writing ability. The District’s placement included classes, albeit small special education classes, where he could develop intellectual skills requiring synthesis, especially verbal reasoning and comprehension in both speaking and writing. This record establishes that the District’s proposed placement was reasonably calculated to provide educational benefits for Gregory and was therefore an “appropriate” public education under the Act.

VII

REIMBURSEMENT, COSTS, AND ATTORNEYS’ FEES

Where a school district improperly changes a special education placement or proposes an inappropriate placement, courts may order the district to reimburse parents who pay to maintain an appropriate placement. Burlington School Committee v. Department of Education, 471 U.S. 359 , 105 S.Ct. 1996, 2003 , 85 L.Ed.2d 385 (1985). We have held that the District’s proposed placement was “appropriate” and that the private tutoring Gregory’s parents paid for was not special education pursuant to an IEP. We therefore reverse the trial court’s order that the District reimburse Gregory K.’s parents for their tutoring expenses.

The parents are no longer the prevailing party under Fed.R.Civ.P. 54(d), so we also reverse the trial court’s grant of costs. Not having prevailed, the parents are not eligible for attorney’s fees. See Handicapped Children’s Protection Act of 1986, Pub.L. 99-372 § 2 , 100 Stat. 796 (1986) (to be codified at 20 U.S.C. § 1415 (e)(4)(B)).

VIII

MANDATORY REASSESSMENT

The trial court denied the District’s post-trial motion to compel testing of Gregory. The District argues that it must reassess Gregory “at least once every three years.” WAC 392-171-516. If the parents want Gregory to receive special education under the Act, they are obliged to permit such testing. See DuBois v. Connecticut State Bd. of Ed., 727 F.2d 44 , 48 (2d Cir.1984); Carroll v. Capalbo, 563 F.Supp. 1053, 1058 (D.R.I.1983). If the parents wish to maintain Gregory in his current private tutoring program, however, the District cannot require a reassessment.

CREDIT FOR TUTORING

In its order of May 16, 1986, the trial court ordered the District to “issue and record full scholastic credit for all work completed by Gregory K. under the tutelage of Lois Lewis after February 11,1983, to the present day.” The District, not the court, decides this question, unless the court can determine from the record that the District has varied from its usual standards in denying such credit. Nothing in the record supports such a finding. We therefore reverse the order.

REVERSED.

. The correct scaled score for this test may have been 80 rather than 85. See notes 2 and 3 below for a more complete discussion of evidence in the record concerning Gregory’s IQ testing. . This was also the only time Gregory was given the Stanford-Binet Intelligence Test. It was given in 1974, when Gregory was six. The school psychologist evaluating the results noted at the time that they were "well within the Low Average ranges of ability." The "derived score” computed from the results was 85. Unrebutted testimony by one of the District’s psychologists indicated that out-of-date norms had been used in computing the score. Using the correct norms would have produced a derived score of 80. . Subsequent IQ testing of Gregory was done chiefly by using the Wechsler Intelligence Scale for Children (WISC), though somewhat different versions were used at different times. In 1975, Gregory was given a full WISC, resulting in a scaled score of 70. In 1976, he was given a partial WISC, resulting in a prorated score of 69. In 1979, he was given a full WISC (revised format), resulting in a scaled score of 69. In 1982, Gregory was given the Comprehensive Tests of Basic Abilities, which produces, among other results, a “cognitive skills index” that is comparable to scaled IQ scores. His CSI was 69. In 1983, Gregory was again given a full WISC (revised format), resulting in a scaled score of 66.