Artificial neural networks are widely used in image classification tasks. The reliability of examination set image classification by individual three-layered neural network can vary in great diapason. To achieve examination set classification reliability’s stability neural network committees are used. In this work, a new neural network committee method is proposed. The proposed method is experimentally compared with the existing methods (plurality voting algorithm, weighted voting algorithm, ensemble average algorithm). The experiments were conducted in two areas: criminal and medical. In criminal area, photographs of human faces were used to test the proposed method. To complicate the task, faces of twins, which are hard to distinguish for by human eyes, were used. In medical area of the experiment, medical images of a brain, acquired with help of computer tomography or magnetic resonance were used. Experiments show that the proposed method has advantages over the known methods in the task of face recognition. The proposed algorithm showed greater reliability in classification during all time of neural network learning. In the task of medical image recognition the proposed method showed results that did not differ from the existing methods. This result can be connected with the need to segment the pathology zone on the image.