A method for analyzing single flank test results is proposed. The analysis yields geometrical deviations of spiral bevel gears. It is based on extracting information on the meshing tooth surfaces from the transmission error curve and digital photograph of the contact pattern. A neural network is trained by a batch of computer simulated transmission error curves and respective contact patterns belonging to systematically varied geometrical deviations. Taking advantage of the generalization capability of the neural network, it can be used to yield tooth flank topography errors on the basis of single flank test measuring results. Giving similar results to coordinate measurement, the method extends the capabilities of single flank test significantly. It saves costly coordinate measuring time, which may be especially advantageous in case of large bevel gears. It takes only minor changes for the analysis to be applicable for any types of gears.