🏛️ Issues and Advocacy Training
Courses › Surveillance Technology and Privacy

Lesson 2 of 9 · about 15 minutes

How License Plate Readers Work

You'll be able to

  • Describe how an automated license plate reader captures and stores a read
  • Explain hotlists, alerts and why officers must confirm a hit
  • Estimate how many reads a camera network stores from traffic counts and retention
  • Name the main vendors and how their sharing networks differ from a single agency's data

A plate reader is a camera with a narrow job. It photographs every vehicle that passes, reads the plate with software, and writes down where and when. It does this all day, for every car, whether or not anyone is looking for that car. That last part is the whole debate in one sentence.

From photo to record

An automated license plate reader, or ALPR, usually pairs a high-speed camera with infrared lighting so it can read plates at night and through glare. Software finds the plate in the frame and runs optical character recognition to turn the image into text, like "7KRT219." Many newer systems also tag the vehicle's make, color, body type and features like a roof rack or a bumper sticker. That means a search can find a car even when the plate was unreadable.

Each pass creates a read. A typical read holds:

  • the plate text and the state, if the software can tell
  • one or more photos of the vehicle, often showing part of the street
  • date and time to the second
  • camera location and direction of travel
  • a confidence score for the plate text

Cameras come in three main forms. Fixed cameras sit on poles, overpasses and highway ramps. Mobile cameras ride on patrol cars and read parked cars as an officer drives by, which lets an agency sweep a whole neighborhood in a shift. Trailers get towed to a spot and left there for weeks.

Camera Plate OCR Hotlist check Alert to officer Database (retention clock) Sharing match every read

Hotlists and alerts

Every read is compared to one or more hotlists. These can include the national stolen vehicle file, plates tied to warrants, Amber and Silver alerts, and lists an agency builds itself, like a suspect in a burglary series. A match sends an alert to dispatch or to officers' phones and laptops.

Here is the catch. The software misreads plates. An 8 can become a B, a state can be guessed wrong, a hotlist entry can be stale because the stolen car was recovered last month. EFF documents a 2009 San Francisco case in which a misread led officers to hold a woman at gunpoint over a car that was not stolen, and a Colorado case in which a family was detained after a hit on the wrong vehicle. Good policies require officers to confirm the plate and the state with their own eyes, and confirm the hotlist entry is current, before a stop. Ask whether your local policy says that.

The database is where privacy lives

An alert happens in real time and concerns a car someone was already looking for. The stored reads are different. They let an investigator search backward: where was this plate last Tuesday, what cars passed the corner near the burglary between 2 and 3 a.m., which vehicles show up near both crime scenes. That search can include the cars of people nobody suspected when the reads were taken.

So the retention period is the number to find first. State laws range from three minutes in New Hampshire to 21 days in Maine and Virginia, 60 days in Minnesota, nine months in Utah and 18 months in Vermont, according to the National Conference of State Legislatures as of September 2026. Many states set no limit at all, and then the vendor's default or the agency's policy controls. Flock's commonly reported default is 30 days. Some Idaho police departments told the Idaho Capital Sun in August 2026 they keep data 30 days for that reason.

Worked example: how big is the database?

The invented town of Cedar Bend installs 18 fixed cameras on its main roads. City traffic counts show the average camera sees 7,500 vehicles a day.

  • Reads per day: 18 x 7,500 = 135,000
  • Stored at 30-day retention: 135,000 x 30 = 4,050,000 reads
  • Stored at one-year retention: 135,000 x 365 = 49,275,000 reads

Now apply EFF's finding from 63 California agencies that only about 0.05 percent of reads matched a public safety interest at capture. 0.05 percent of 135,000 is 0.0005 x 135,000 = 67.5, so roughly 68 reads a day would be hotlist-relevant. The other 134,932 are ordinary drivers. Supporters of the systems point out that some of those ordinary reads later become the clue that solves a hit-and-run. Both facts are true. An honest description of the system, from either side, uses both numbers.

Now suppose a 1 percent misread rate (a made-up figure for this example; ask your agency for its own). That would be 1,350 wrong plate texts a day. Most never matter. The ones that match a hotlist by accident are the ones that lead to bad stops.

Vendors and networks

The two names you will see most are Flock Safety and Vigilant Solutions, a subsidiary of Motorola Solutions. Others include Rekor, Axon, Elsag and Jenoptik. Vigilant's sister company, Digital Recognition Network, gathers plate reads from cameras on repossession company vehicles, which means some plate data is collected by private companies and then sold or made available to police.

Flock contracts are commonly written as subscriptions: the agency pays per camera per year rather than buying hardware outright. Check the price sheet in your agency's contract instead of relying on figures from other cities. Flock also sells to homeowners associations and businesses, and those private customers can choose to share with police.

The vendor platform makes sharing easy. An agency can let neighboring departments search its data, join a statewide group, or join a national lookup. Flock's trust page says agencies control their own data, every search is logged and tied to a user, and nothing is shared automatically. That puts the decision, and the paper trail, with the local agency. California's law, S.B. 34, allows public agencies to share ALPR data only with other public agencies. Lesson 3 covers what happened when out-of-state and federal users searched data in states that restricted it.

SettingWhy it mattersRecord that shows it
Retention daysHow far back searches can reachPolicy, contract, vendor settings screenshot
Sharing partnersWho else can search local dataSharing list or "organization access" report
Hotlist sourcesWhat triggers stopsPolicy, hotlist entry logs
Search purpose fieldWhether searches can be auditedAudit log export
Private camera linksWhether HOA or business cameras feed policeContract, integration list
The camera count gets the attention. The retention period, the sharing list and the audit log tell you what the system can actually do.

Common mix-ups

  • Claim: Plate readers only record criminals' cars. What the evidence shows: They record every passing vehicle. Only the hotlist check is targeted.
  • Claim: Plate readers use face recognition. What the evidence shows: Most ALPR products, including Flock's per its trust page, do not identify faces. They do photograph vehicles, and sometimes occupants are visible in the image.
  • Claim: The city owns all the data, so the vendor has none. What the evidence shows: Ownership language varies by contract. Read the data ownership, data use and termination clauses.

Reporting angle

Get the contract and the council or commission minutes approving it. Find the per-camera price and the total. Ask for the camera location list (some agencies withhold it, citing safety, so ask for at least the count by area). Ask for the retention setting, the sharing list, and a 30-day export of hotlist alerts with outcomes: how many alerts, how many confirmed, how many stops, how many arrests. That last figure is the one most agencies have never calculated.

Check yourself

1. What is a hotlist in an ALPR system?

2. A network of 25 cameras each reads 6,000 vehicles a day and keeps data 30 days. About how many reads are stored at any time?

3. Why can the retention period matter more than the number of cameras for privacy?

4. According to EFF's 2021 review of 63 California agencies, roughly what share of ALPR data was relevant to a public safety interest when captured?

5. Which statement about sharing networks is accurate?

Terms in this lesson

Automated license plate reader (ALPR)
A camera system that photographs vehicles, uses optical character recognition to read the plate, and stores the plate, time, location and image.
Hotlist
A list of plates of interest loaded into an ALPR system. A read that matches triggers an alert.
Read
One record of one vehicle passing one camera: plate text, photo, time, place and often vehicle details.
Retention period
How long stored reads are kept before automatic deletion.
Sharing network
An arrangement that lets agencies search each other's ALPR data, often through the vendor's platform.

Full glossary

Sources