Please see MSE criticism part at this page.

http://en.wikipedia.org/wiki/Mean_squared_error

I use RMSE calculations between two raster datasets containing Z values in ArcGIS.

(free toolbox: http://resources.arcgis.com/gallery/file/geoprocessing/details?entryID=59101EDB-1422-2418-88CF-7D266D2881AC).

For my case RMSE is ok, since it required for input data specifications and raw data valdiation, for other domains - may be questioned.

Thank you very much Andrius. I read MSE criticism, but it about the bias MSE or RMSE can have. All this aspect are too specific (but very useful as well for me) for what I meant in the question. Anyway I think you a basically agree with that specify accuracy in term of error (MSE, RMSE or absolute mean error) is equivalent to specify it in term of (1 - RMSE percent) for example.

The RMSE is an estimate of the standard deviation of your data. A theoretical lower bound on the RMSE can be computed using the Cramer-Rao Lower Bound. If the distribution of the data is Gaussian, and you have enough data, it is a very good approximation of the standard deviation. It is not very good if your data is non-Gaussian and contains any outliers.

The RMSE is the square root of the variance of the residuals. It indicates the absolute fit of the model to the data–how close the observed data points are to the model’s predicted values. Lower values of RMSE indicate better fit. RMSE is a good measure of how accurately the model predicts the response, and it is the most important criterion for fit if the main purpose of the model is prediction.

Hi Ugo

RMSE measures percentage of errors. It is a good accuracy indication. You can say RMSE of your model is x or say the accuracy of your model is (1 - RMSE%).