Describes a method for the automatic detection of small dents in car bodies which are not visible until the paintwork. For automatic error detection, two problems have to be solved. The accuracy of the measurement system has to be on a sufficient level and the errors have to be detected in the 3D measurement data. The measuring of the surface shape can be done by an optical 3D measurement system. This system consists of two cameras and one projector. The problem of error detection is solved by a method based on neural networks. The measurement data of one or more master workpieces is stored in the weights of a neural network. The calculation of the difference between the measurement data and the output of the neural network gives the resulting error surface. In the paper, a combination of both technologies is described. This dent detection method is illustrated by an example.