How GPS Receiver Accuracy Is Specified and What That Means for Location Evidence

Published on May 7, 2026

Contact Mark CV Download

GPS “accuracy” claims are not directly comparable unless you know the statistic and the test conditions

When a GNSS or GPS receiver sits still, it does not report one fixed point. It reports a cloud of points that move because satellite timing and orbit error, signal bounce off surfaces, and the ionosphere and weather all shift the solution.

That cloud can be tight or wide, and it can also sit off to one side of the true point. I treat “accuracy” as how close the cloud sits to the true point, and I treat “precision” as how tightly the points cluster and how often they land near the center.

Manufacturers still publish one number, but they do not always mean the same thing. One datasheet may state 1 m CEP, another may state 1 m 2DRMS, and a third may state 1 m RMS or R95. Those labels hide different confidence levels, so equal meters can reflect different certainty. Test setup also shifts the result.

A vendor may compare results to a surveyed marker, to a prior average, or to another reference, and that choice changes what “true” means for the claim.

Vertical accuracy adds another gap because many specs omit it, and in many practical setups height error exceeds horizontal error, so a horizontal spec does not support the same radius in elevation. To compare claims, you first need the statistics behind the number.

What the common accuracy statistics mean (CEP, RMS or DRMS, 2DRMS, R95) and the confidence they imply

All of the common spec terms describe a radius around a true point, but each term uses a different probability target. CEP, circular error probable, is the radius that contains 50 percent of fixes in the horizontal plane.

If a receiver claims 1 m CEP, it means half the fixes fell within 1 meter of the true point in the stated test.

RMS, root mean square, describes average squared error, and vendors often publish horizontal RMS or DRMS, distance RMS, for two dimensions. In practice DRMS corresponds to about 63 to 68 percent of points, and the exact probability depends on how similar the east and north errors are.

2DRMS is two times DRMS. Many sources associate it with about 95 to 98 percent containment for horizontal error, again with the range driven by how circular the error cloud looks.

R95 is the radius that contains 95 percent of points by definition.

The key idea for evidence work is simple: a “1 m CEP” claim does not match a “1 m 2DRMS” claim. A CEP number targets the middle of the distribution, while 2DRMS and R95 target the tails where court questions often sit.

Once you know what probability the spec implies, you can ask whether the test conditions match your case conditions.

Mechanisms that make the spec diverge from real-world location evidence: environment, corrections, and non-Gaussian error behavior

Receiver fixes wander because errors change over time, and the environment shapes how large those errors get. Open sky often yields tighter scatter than canopy, streets with tall buildings, vehicle interiors, or near reflective surfaces.

Multipath, where signals bounce before they reach the antenna, can push points in one direction and create outliers that do not look like “typical” error.

Correction mode also matters. Autonomous positioning uses only satellite signals. SBAS or other differential corrections often shrink error, and RTK can shrink it further when it holds a fixed solution. If you compare a spec that assumes corrections to data collected without them, you misstate what the number supports.

Many published measures assume errors follow a normal, bell-shaped distribution. Real error often violates that assumption, especially over short sessions or in difficult environments, and studies that use large real-world datasets report that a 2DRMS-based estimate can differ from the empirical 95 percent radius.

That gap matters when you translate “accuracy” into “within X meters.” Those limits set up the next question: what can a spec justify in testimony about a specific fix or track?

Implications for location evidence: what an accuracy statement does and does not support

An accuracy spec gives you a probability statement tied to conditions, not a guarantee for a single fix. If you say “within 1 meter” without stating whether that means 50 percent, 95 percent, or another level, you hide the core uncertainty.

I try to phrase it as “under stated conditions, the manufacturer reports that about N percent of fixes fall within R meters,” and then I ask how far the case conditions depart from the test.

You also need to separate receiver output from map or GIS output. A phone may record locations, and a mapping tool may display them, but the map layer itself carries its own error, update limits, and opaque processing.

Courts and commentators have raised concerns about treating tools like Google Maps or Google Earth as unquestionable sources without authentication, because data sources, algorithms, and updates can introduce errors that do not appear on the screen.

Finally, GPS-based location differs from historical cell-site location. Cell-site records often show the first and last serving cell sector for a call, and overlap, obstruction, and network load can drive tower choice, so the implied area can span many blocks or more.

This leads to a practical need: a short, repeatable way to compare receiver claims and state a defensible radius.

Minimal checklist for comparing receivers and framing a defensible location-radius statement

A few inputs are worth gathering before translating a datasheet number into a location statement. The metric and its confidence level matter because CEP, RMS or DRMS, 2DRMS, and R95 do not reflect the same probability.

The correction mode that applied during data collection (autonomous, SBAS or other differential, or RTK) can place the receiver in a different accuracy band than the published spec.

Environment conditions such as open sky, canopy, dense urban, or indoor use affect how much the fixes scatter, and horizontal and vertical error are worth keeping separate unless there is support for both.

Where a receiver point is compared to a surveyed point or another dataset, both should share the same datum and coordinate reference frame so that a reference mismatch is not mistaken for a positioning error.

GIS or mapping outputs are generally treated as separate evidence requiring their own foundation, with disagreements addressed as weight issues unless corroborated by an independent source.

With those inputs in hand, a datasheet number can be translated into a probability statement grounded in the conditions that actually applied.

Contact Mark CV Download