In this paper a face recognition algorithm based on multiscale wavelet representations and model adaptation is proposed. The models used in this work are from Linear Associative Memory method and fast compensated in simulated testing phase by adaptively learning from the given simulated testing data. The proposed adaptation algorithm is incremental. It has low time and space complexity. Through compensating models with simulated testing data, this method can efficiently reduce the mismatch between training and testing data, substantially improving the performance of classifier. The new recognition method was tested using two widely used face datasets, including MIT-CBCL face database and Olivetti Research Laboratory (ORL) face database. Results indicate that our algorithm is effective and duo to the computational simplicity, our algorithm is also efficient.