dissenting as to Parts 111(A) and (C).

Fifty million vehicles a year pass through the Temecula and San Clemente checkpoints, which straddle the two major arteries connecting San Diego and Los Angeles about 70 miles inside U.S. territory. It’s quite an experience: Every car, van, truck and coach heading north is slowed to a crawl so uniformed law enforcement officers can peer inside for ... you’re not exactly sure what. If the officers think you’re worth investigating, they pull you over, question you and look *1316 inside your car. They alone decide whether a more thorough search will follow.

In United States v. Martinez-Fuerte, 428 U.S. 543 , 96 S.Ct. 3074 , 49 L.Ed.2d 1116 (1976), the Supreme Court approved such internal checkpoints as a necessary means of stemming the flow of undocumented aliens. Experience has shown, however, that the checkpoints yield a far richer harvest — a cornucopia of contraband, particularly illegal drugs. 1 Perhaps this is an accident — -just a lot of lucky breaks on the part of the Border Patrol. But it may be much more: There’s reason to suspect the agents working these checkpoints are looking for more than illegal aliens. If this is true, it subverts the rationale of Martinez-Fuerte and turns a legitimate administrative search into a massive violation of the Fourth Amendment.

A. Administrative searches are an oddity of Fourth Amendment jurisprudence. They involve judicially-approved searches and seizures of countless people and places, with no particularized suspicion and normally without a warrant. To justify such a wholesale exemption from ordinary Fourth Amendment requirements, the government must establish two critical elements: first, that there’s a compelling need for the intrusion; and second, that the intrusion is strictly limited to fulfilling that need. United States v. $121,-570 U.S. Currency (Campbell), 873 F.2d 1240, 1244-45 (9th Cir.1989).

Security checks at airports and public buildings, 2 sobriety checkpoints, 3 safety inspections of private homes and businesses, 4 home visits by welfare caseworkers, 5 agricultural quarantine checks at airports, 6 investigations following a fire, 7 drug testing of certain public employees, 8 unannounced inspections of businesses in closely regulated industries 9 — all of these and many more have been upheld under an administrative search rationale. What this means is that government officials routinely invade the privacy and property of countless millions; hardly anyone escapes their clammy grasp.

With this power comes a vast potential for abuse. The Supreme Court has therefore “repeatedly emphasized the importance of keeping criminal investigatory motives from coloring administrative searches.” Id. at 1244 . We too have stressed the need to keep these searches from becoming “infected by general law enforcement objectives, and the concomitant need for the courts to maintain vigilance.” Id.; see Davis, 482 F.2d at 909. These judicial admonitions are fueled by the fear that, if not carefully circumscribed and monitored, administrative searches can destroy the promise of the Fourth Amendment.

Consider building inspections. A city may — without particularized suspicion — send inspectors into private homes to check for housing code violations. Camara, 387 U.S. at 538, 87 S.Ct. at 1735. It may do so, *1317 however, only if “the inspections are neither personal in nature nor aimed at the discovery of evidence of crime_” Id. at 537, 87 S.Ct. at 1735. Suppose one sunny day the City of Los Angeles decides to give its building and safety inspectors general law enforcement training. As part of this program, they are taught the smell of certain illegal drugs, how to spot illegal weapons and illicit gambling operations. Each inspector is given a buddy on the police force whom he is to call if he finds evidence of criminal activity while rummaging through people’s homes. The city rewards and promotes inspectors for assisting with police investigations. Such a program would make a lot of sense from a law enforcement perspective, but it would surely be incompatible with the Supreme Court’s rationale in Camara.

Of course, if a safety inspector happens to see a dead body in someone’s living room during a routine inspection, it’s fine for him to tell the police. But the closer the working relationship between the police and the inspector, and the more the inspector sees sleuthing as part of his mission — as a way to advance his career — the more he’ll be tempted to stretch his inspection beyond what safety alone calls for. The inspector may then be just as likely to conduct a “routine” building inspection not because of safety code procedures or tenant complaints, but to help the police figure out what’s going on inside.

Or take the case of sobriety checkpoints. In Michigan v. Sitz, 496 U.S. 444 , 110 S.Ct. 2481 , 110 L.Ed.2d 412 (1990), the Supreme Court held that the police can stop traffic on public streets and highways, observe drivers and briefly question them. Those who look intoxicated are then pulled over and tested.

Suppose the California Highway Patrol and the DEA hatch a plan to use drunk-driving checkpoints in the war against drugs: CHP officers agree to set up the checkpoints at freeway entrances the DEA believes are frequented by drug couriers, and the DEA gives the CHP officers drug courier profiles and trains them to scan drivers and passengers. Suppose there is also a tacit understanding that, in making the highly discretionary decision about whom to pull over and test, CHP officers consider whether anyone in the car fits a DEA profile, whether the ear is popular with drug dealers, and a variety of other factors that point to drug dealing. While these sobriety stops might look exactly like those approved in Sitz , the effort to use them for unrelated law enforcement purposes would fatally undermine the Supreme Court’s rationale in that case.

That this is more than idle speculation is the lesson of United States v. $124,570 U.S. Currency (Campbell), 873 F.2d 1240 (9th Cir.1989). The case involved a bundle of cash found during a search of an airline passenger’s hand luggage. We had previously approved such security checks under an administrative search rationale, subject to the express condition that they not be used for general law enforcement purposes. Davis, 482 F.2d at 911-12.

The U.S. Customs Service and the Seattle Port Police nevertheless adopted a program of rewarding Flight Terminal Security (FTS) agents for finding drugs or large sums of money, and the FTS adopted a policy of reporting such finds. Campbell, 873 F.2d at 1245 & n. 3. We held that the administrative search had been tainted:

Id. at 1245 . We noted the FTS’s policy would “very likely influence FTS officers to conduct more searches, and more intrusive searches, than if they focus[ed] on air safety alone.” Id. The melding of these interests, we held, “effectively transform[ed] a limited check for weapons and explosives into a general search for evidence of crime. This substantially erod[ed] the privacy rights of passengers who travel through the system.” Id. at 1246 .

Campbell illustrates that administrative searches are to law enforcement what a pot *1318 of maple syrup is to honeybees. In earnest pursuit of their mission — and probably with no intention of stepping beyond the bounds of propriety — law enforcement officials will try to get as much as they can from the permitted intrusion. The immigration checkpoints at Temecula and San Clemente are as susceptible to this abuse as the security checks in Campbell. As we noted there, “a sieve through which pass the contents of billions of satchels, purses, briefcases and pockets will naturally strain out much that is of interest to law enforcement.” Id. So too it is with the contents of tens of millions of vehicles that travel from San Diego to Los Angeles.

B. Martinez-Fuerte approved immigration checkpoints for a very narrow purpose— detecting, and thereby deterring, illegal immigrants. But, as in Campbell, the record here suggests that these checkpoints may also be serving broader law enforcement objectives.

Consider just how knowledgeable about drugs were the Border Patrol agents in this case. Agent Van Gorder, who sent Soyland and Whitaker to secondary, testified that her Border Patrol training included cross-training by DEA and Customs. RT 1/25/91 at 33. And that training seems to have paid off: Illegal drags were found in one of every eight cars she had referred to secondary. Id. at 29. Agent Boubel, who conducted the search and arrested Soyland and Whitaker, had been trained to recognize whether someone is under the influence of drugs; he’d also been taught a lot about methamphetamine— its odor, how it’s manufactured and how it’s likely to be transported. Id. at 88-89. While at Temecula, Agent Boubel had smelled methamphetamine over a hundred times and found it every time he’d smelled it in a car. Appendix, infra, at 1324. He also recognized vitamin B powder as a substance commonly used to dilute narcotics. Id. at 1321.

Consider also what happened to Soyland and Whitaker once they got to secondary. The immigration check of their vehicle was brief, which is not surprising given that a Mazda RX-7 is hardly suitable for transporting aliens; Agent Boubel didn’t even bother verifying Soyland and Whitaker’s citizenship. Instead, he spent most of his time conducting what he himself described as a drug investigation, id. at 1321 — examining such things as the contents of a plastic bag, a cigarette pack and the pockets of a jacket, id. at 1321-1322. In Agent Boubel’s own words: “I had smelled the smell that I identified as methamphetamine. I had seen the package on the floor that, in my experience, is consistent with other drug apprehensions that I have made at the checkpoint. So, at that point, I decided that I would ask [Whitaker] for permission to search the interior of the vehicle for any possible illegal drugs.” Id. at 1321. 10

Now if a Border Patrol agent stumbles across a brick of marijuana as he’s checking for aliens, it would be perfectly okay for him to turn the occupants over to the DEA, just as it would be fine for the building inspector to report the cadaver in the living room. But it’s a different matter when the Border Patrol agent conducts a separate investigation after he’s established there are no immigration law violations, particularly if he’s been trained to detect the smells of various drags, instructed to follow his nose to their origin and given an incentive to do so. And it’s a different matter still if the agent who makes the infinitely malleable judgment about whom to pull over to secondary is thinking not just about aliens, but also about drugs, guns, uninspected produce or other contraband he might find. Discovery of such contraband then is no longer an accident; it is a systematic wrenching of the administrative search from its constitutional roots. 11

*1319 My colleagues look at the record here and find my concerns “utterly without support.” Majority at 1314. Reasonable minds can differ, I suppose, but I have much difficulty reading the testimony of Agent Boubel, for example, and concluding that looking for drugs wasn’t a big part of the job as he understood it. His search for immigration violations was entirely perfunctory. See Appendix, infra, at 1320-1321. But one is thoroughly impressed by his dogged determination in pursuing the smell of methamphetamine, his eagerness to spot drug-dealing activity from clues most people would look right past, his willingness to arrest the suspects for three marijuana cigarettes, id. at 1327 — a quantity so small the State of California treats its possession like a minor traffic infraction. Cal.Vehicle Code § 23222. 12 Such a gung-ho attitude is commendable in public servants, but not in conducting administrative searches.

C. “Liberty — the freedom from unwarranted intrusion by government — is as easily lost through insistent nibbles by government officials who seek to do their jobs too well as by those whose purpose it is to oppress; the piranha can be as deadly as the shark.” Campbell, 873 F.2d at 1246 (footnote omitted). As federal judges we have a duty to safeguard the constitutional rights of our citizens from subtle as well as blatant overreaching by government officials. Where the record in a particular case, along with our past experience, see note 1 supra, suggests that law. enforcement officials may be routinely overstepping their authority in conducting administrative searches, we have a special responsibility to inquire further. See United States v. Redondo-Lemos, 955 F.2d 1296, 1298 (9th Cir.1992).

I would therefore remand the case against Whitaker for a factual inquiry into whether the government officials who operate the Temecula and San Clemente checkpoints have remained faithful to their original purpose. This wouldn’t be a particularly burdensome inquiry: It would involve only an objective examination of the operation of these checkpoints, not the motives of any particular official, making it far less intrusive than the inquiry we approved in Redondo-Lemos . The key question would be whether the checkpoints are being run as contemplated by Martinez-Fuerte , or whether the policies, programs, directives and incentives put in place by the government, or any customs and practices that have developed with the government’s tacit approval, have toned Temecula and San Clemente into general law enforcement checkpoints. 13

*1320 Given the strong hints that the Constitution is being routinely violated at these checkpoints, we owe it to ourselves and the public we serve to look into the matter. Even without an order of this court or the district court, the Department of Justice would be well-advised to establish the bona fides of these checkpoints by presenting such evidence to the district court on remand. “[T]he necessity to preserve individual rights ... in the course of investigations and prosecutions” is, after all, not the judiciary’s burden alone. Address by Attorney General Janet Reno to Justice Department Employees, Fed. News Serv., Apr. 2, 1993.

APPENDIX

Excerpts from Testimony of United States Border Patrol Agent Ralph Boubel, 1/25/91, at 62-96.

Direct Examination:

Cross-Examination:

. See, e.g., in 1992 alone, United States v. Reinhardt, 980 F.2d 740 (9th Cir.1992) (table); United States v. Reynolds, 979 F.2d 857 (9th Cir.1992) (table); United States v. Koshnevis, 979 F.2d 691 (9th Cir.1992); United States v. Williams, 977 F.2d 594 (9th Cir.1992) (table); United States v. Morales, 972 F.2d 1007 (9th Cir.1992); United States v. Bowers, 967 F.2d 592 (9th Cir.1992) (table); United States v. Cortes, 967 F.2d 592 (9th Cir.1992) (table); United States v. Preciado-Robles, 964 F.2d 882 (9th Cir.1992); United States v. Ponce-Rodriguez, 963 F.2d 381 (9th Cir.1992) (table) — all drug cases coming from an immigration checkpoint.

. United States v. Davis, 482 F.2d 893 (9th Cir.1973); McMorris v. Alioto, 567 F.2d 897 (9th Cir.1978).

. Michigan Dep't of State Police v. Sitz, 496 U.S. 444 , 110 S.Ct. 2481 , 110 L.Ed.2d 412 (1990).

. Camara v. Municipal Court, 387 U.S. 523 , 87 S.Ct. 1727 , 18 L.Ed.2d 930 (1967); See v. City of Seattle, 387 U.S. 541 , 87 S.Ct. 1737 , 18 L.Ed.2d 943 (1967).

. Wyman v. James, 400 U.S. 309 , 91 S.Ct. 381 , 27 L.Ed.2d 408 (1971).

. United States v. Schafer, 461 F.2d 856 (9th Cir.1972).

. Michigan v. Tyler, 436 U.S. 499 , 98 S.Ct. 1942 , 56 L.Ed.2d 486 (1978).

. Skinner v. Railway Labor Executives' Ass'n, 489 U.S. 602 , 109 S.Ct 1402 , 103 L.Ed.2d 639 (1989); National Treasury Employees Union v. Von Raab, 489 U.S. 656 , 109 S.Ct. 1384 , 103 L.Ed.2d 685 (1989).

. New York v. Burger, 482 U.S. 691 , 107 S.Ct. 2636 , 96 L.Ed.2d 601 (1987) (auto junkyards); Donovan v. Dewey, 452 U.S. 594 , 101 S.Ct. 2534 , 69 L.Ed.2d 262 (1981) (underground and surface mines); United States v. Biswell, 406 U.S. 311 , 92 S.Ct. 1593 , 32 L.Ed.2d 87 (1972) (firearms and ammunition dealers); Colonnade Catering Corp. v. United States, 397 U.S. 72 , 90 S.Ct. 774 , 25 L.Ed.2d 60 (1970) (liquor licensee).

. Excerpts of Agent Boubel's testimony are reproduced in the Appendix to this dissent. Interested readers can make up their own minds whether this is someone who considers drug inspections a routine part of his job.

. An explicit policy may not be necessary to taint these checkpoints. It may suffice if there is a tacit policy, if Department of Justice officials in charge of the checkpoints are aware that field agents regularly use the checkpoints for general law enforcement. See City of St. Louis v. Praprotnik, 485 U.S. 112, 130 , 108 S.Ct. 915, 927 , 99 L.Ed.2d 107 (1988) (plurality opinion) (potential section 1983 liability “if a series of decisions by a subordinate official manifested a 'custom or usage' of which the supervisor must have been aware"); Gillette v. Delmore, 979 F.2d 1342, 1349 (9th Cir.1992) (custom or informal policy may be proven through "evidence of repeated constitutional violations for which the errant municipal officials were not discharged or repri *1319 manded”); Oviatt by and through Waugh v. Pearce, 954 F.2d 1470, 1477 (9th Cir.1992)-(poIi-cymaker’s inaction in face of problem constituted policy for purposes of section 1983 liability); see also Grandstaff v. City of Borger, 767 F.2d 161, 171 (5th Cir.1985) (prior incidents of misconduct tend to prove a pattern or custom and policymaker’s accession to it). While these cases arise in a somewhat different context, they provide an apt analogy to our situation. Because administrative searches are so easily diverted from their narrowly defined purposes, government officials have an affirmative responsibility to keep them from being misused. Acquiescence in a custom or practice that routinely disregards the limits of particular administrative searches might itself be sufficient to establish a breach of this responsibility.

. That Agent Boubel considered drug violations not all that different from immigration violations might be gleaned from the following exchange on cross-examination:

Appendix, infra, at 1326.

. While we should rely on the parties and the district court to frame the precise questions on remand, the following would no doubt be relevant: Are the checkpoints staffed entirely by Border Patrol and other INS agents? If not, what other agents are normally present, in what numbers and for what purpose? Do the Border Patrol agents at the checkpoints get trained by the DEA and other law enforcement agencies? If so, is this the training given to Border Patrol agents generally, or is it checkpoint-specific? Do the agents get regular alerts from other law enforcement agencies as to what (other than aliens) to look for? Are the agents trained to recognize drug courier profiles and do they take these into account in deciding when to pull cars over to secondary and how thorough a search to conduct? Are the agents instructed to stop their inspection once they’re satisfied there are no aliens, or are they encouraged to go further if they think they might find other contraband? Are the checkpoints regularly equipped for investigations into other contraband? See, e.g., United States v. Taylor, 934 F.2d 218, 219 (9th Cir.1991) *1320 (San Clemente Border Patrol agent used dog trained to alert to drugs and people); United States v. Perez, 936 F.2d 581 (9th Cir.1991) (table) (Temecula Border Patrol agent used dog trained to alert to drugs and people). Does the Border Patrol have a policy of rewarding agents who uncover contraband other than aliens? Are the position descriptions, performance goals and other standards for measuring the conduct and performance of Border Patrol agents at Temecu-la and San Clemente keyed to detecting illegality other than aliens? Are agents frequently cross-designated as Border Patrol agents and Customs or DEA agents? See, e.g., Taylor, 934 F.2d at 219 (Border Patrol agent cross-designated as agent of DEA and Customs). In justifying budget and personnel resources for Temecula and San Clem-ente, does the Border Patrol rely on time and resources allocated to look for things other than aliens? Have the Border Patrol agents been adequately trained to respect the limits of these searches and admonished when they do not? See note 11 supra.