Dairy industry has a strong interest in quick and inexpensive techniques able to estimate milk fat content both in off and on-line processes. In the present study, a simple and inexpensive optical technique is preliminary described, discussed, and then tested for the prediction of this parameter. It's essentially based on the different intensity and spectral emission of a tungsten lamp during its progressive lighting combined with the photodiode wavelength sensitivity. Measurements were carried out at 22 degrees C on thirteen samples of commercialized milk (and mixture of them) (fat content ranging from 0.05% to 3.8%). The influence of the milk temperature on the optical behaviour was also investigated. The detected voltage waveform was strictly and not-linearly correlated with the fat content (R-2 up to 0.985) and predictions with R-2 up to 0.997 were obtained by using artificial neural networks (ANN). Milk temperature differently influenced the measurements for milks with various fat content. (C) 2015 Elsevier Ltd. All rights reserved.