Multiple EPSG Transformations Exist — Which One Should You Use?

Look up almost any classical datum on this site and you will find not one transformation to WGS 84 but several — NAD27 has dozens, ED50 and Pulkovo 1942 have country-by-country lists. None of them is "the" transformation. Choosing well takes about thirty seconds once you know what to read.

Why multiple transformations exist

Classical datums were realised by ground triangulation, and the networks bent: systematic distortions accumulated differently in different regions. A single Helmert fit over the whole datum averages those distortions; a regional fit absorbs them. So EPSG publishes one operation per fitted region — "ED50 to WGS 84 (Spain)", "(Norway offshore)", "(Israel)" — each with its own parameters and honesty about accuracy.

The three fields that decide it

1. Area of use. The transformation is only valid inside it. This is the first filter and it is non-negotiable: parameters fitted for Norway encode Norway's network distortions, not Spain's. On every transformation page here, the area of use is listed per operation.

2. Accuracy. EPSG states it explicitly (±1 m, ±35 m…). Among operations whose area covers your data, prefer the best accuracy — but read it as a promise scope too: a ±1 m operation derived from 100 stations beats a ±1 m fit from 4 stations at the edges of its area.

3. Method. Geocentric translations (3-parameter) are the bluntest; position-vector/coordinate-frame Helmert (7-parameter) fit better; grid-based methods (NTv2, NADCON, OSTN15) capture the actual distortion surface and reach decimetres or better. If a grid transformation covers your area and your tooling supports the grid file, it is almost always the right choice.

Practical rules

• Filter by area of use, then sort by accuracy, then prefer grids over Helmert over translations.
• For legal/contractual data (concessions, cadastre), use the operation the authority specifies — even if a "better" one exists.
• Document the EPSG operation code (e.g. EPSG:1133) in your metadata. Two people applying different ED50 transformations to the same data will disagree by metres and both be "right".
• Mind the direction conventions: EPSG defines operations in one direction; applying one backwards without negating properly is a classic metre-scale bug. PROJ handles inversion correctly — hand-rolled code often does not.
• If no operation covers your area, the nearest-region parameters are a guess. Say so in the metadata.

Every transformation page on this site lists all published operations for the pair, sorted by accuracy, with method, parameters and area of use — the three fields above, side by side.