Face identification is a very complex problem to be solved, since it can be used in wide range applications such as identity authentication, access control, surveillance, security, and as part of a robot vision system. This research investigated a possible face identification scheme, preceded by an image pre-processing stage intended to obtain best fitted face data at hand. The feature extraction and data reduction of face images were based on a wavelet transforms. The final steps utilized the power of artificial neural networks, specifically the so called learning vector quantization (LVQ). The face identification scheme explicitly was to classify face expression into six representations (happiness, sorrow, hate, anger, surprise and worry). The results indicate that the use of wavelet transforms and artificial neural network (LVQ) performed sufficiently well. With various process combinations the highest rate success was 79.17%, while the lowest was 52.50%. Included were the addition of noises, image rotations and zooming, as well as the forms of wavelets used (Haar, Daubechies and Coiflet).