Interferometry is widely used in bathymetric applications to relate short time delays between two separate sensors to wavefront arrival angles. The main difficulty of this method concerns its sensitivity with regard to the distance between the two sensor receivers, commonly called baseline. Indeed, the baseline, which is only a few wavelengths long, is sufficient to triangulate the direction of an echo, which can be several hundred meters away. Due to the usual short baseline size, the direction of arrival (DOA) resulting from an interferometric measurement is very sensitive to noise, which can reduce accuracy and handicap the data for immediate use. Interferometric measurements are confined to a -length interval, and yet, the time-delay dynamic range can go beyond this length. As a result, the phase delay is estimated with a -modulo ambiguity. This study focuses on the accuracy problem and tries to make a bathymetric performance estimation. The statistical behavior of the interferometric signal is the keystone of the problem as it allows the bathymetric measurement error to be derived. Because of the very singular nature of the interferometric signal, the noise aspect is not analyzed as a classical additive perturbation, but as a decorrelation between receivers. Thus, a coherent error fusion (CEF) makes it possible to directly integrate the impact of several degrading phenomena into the interferometric-phase probability density function (pdf) and thus, to improve the bathymetric performance prediction compared with a classical root mean squared error (RMSE) integration. Finally, the CEF prediction model is tested on real data collected with a 455-kHz sidescan sonar, and compared with the RMSE-based prediction approach.