In this work, we present a neural networks committee for optical character recognition of seven-segment display digits. The aforementioned digit writing convention restricts the general handwriting recognition problem into a task that can be tackled using an automated approach. Numerous practical applications are available for this type of digit recognition: various forms, written exams etc. We consider three different digit recognition techniques. An appropriate feed-forward neural network is devised for each technique. We use two different methods to determine the optimal topology of neural networks:(1) a traditional, manual approach, and (2) automated topology optimizing system, based on the genetic algorithm. To further improve the performance, a committee of neural networks is developed. In this paper, we elaborate the early results of the character recognition system, based on the devised neural networks committee.