In one embodiment of the present invention, a method is disclosed for determining a difference between a sample position and an in-focus position, as well as a vision inspection system. In a first step image data depicting a sample is captured. Next, a feature set is extracted from the image data. Thereafter, the feature set is classified into a position difference value, corresponding to the difference between the sample position and the in-focus position, by using a machine learning algorithm that is trained to associate image data features to a position difference value.