We present the evaluation of different methods for digit recognition for mobile camera phones. The recognition system follows the typical paradigm of object recognition: a) image segmentation, b) feature extraction, and c) object recognition. The image segmentation is based on a local adaptive thresholding method for separating the digits from the background. Then, 22 features derived from the statistical distribution of points were calculated from the binarized digits. For digit recognition, two minimum distance classifiers were compared: Euclidean and Mahalanobis. The results pointed out that Mahalanobis classifier reached the best performance with 98.9% of accuracy when recognizing single digits and 93.1% when recognizing complete lectures (array of 4 or 5 digits).