Multimodal biometric authentication is gaining more and more attention recently. In , a Reduced Multivariate Polynomial model has been used to combine the biometric classifiers with good performance. Although the model parameter can be solved very quickly, it may not be suitable in case when new registration is a frequent task. Moreover, due to wear and tear of sensors, the distribution of match scores can change with time and hence the performance of a fixed classifier may be affected. In this paper, we introduce an adaptive algorithm to track these changes. The new algorithm is demonstrated to address both the registration and match scores distribution changing problems using two biometrics.