Personal authentication is becoming increasingly important, and on-line signature verification is one of the most promising approaches to the authentication problem. A factor known as intersession variability in signatures causes deterioration of authentication performance. This paper proposes a new algorithm for overcoming this problem. We propose an algorithm that integrates a model parameter updating scheme in order to suppress deterioration in the authentication system. A model is provided for each user to calculate the score using fused multiple distance measures with respect to previous work. The algorithm consists of an updating phase in addition to a training phase and a testing phase. In the training phase, a model is generated via Markov Chain Monte Carlo for each individual. In the testing phase, the generated model determines whether a test signature is genuine. Finally, in the updating phase, the parameters are updated with test data by using a Sequential Monte Carlo algorithm. Several experiments were performed on a public database. The proposed algorithm achieved an EER of 6.39%, using random forgery for training and skilled forgery for testing.