Composting is one of the most efficient ways of managing municipal sewage sludge. Recently, due to the increased demand for composting, the issue of conducting this process in cost effective way is of particular importance. Determining the early maturity stage of the composted material can significantly improve the efficiency of surface management of relatively expensive compost plant. The following research presents classification of neural models for determining the early stage of composted mixture of sewage sludge and rapeseed straw, basing on information contained in images of material samples obtained with UV-A illumination. The topology of the MLP network was used in the construction of classification models. As input variables, 25 color parameters and 21 texture parameters were originally used, but also steps were taken to eliminate their number. The classification error for the developed neural models ranged from 1.83 to 4.27%. The best model in terms of the lowest value of error, and the smallest number of input variables required, included 16 neurons in the input layer, 50 neurons in the hidden layer and 2 neurons in the output layer. The following model is characterized by a slightly lower classification error and a slightly simpler structure than the best possible model developed in earlier studies for visible light illumination.