Adaptive models are methods for updating a regression model that enable a soft sensor to follow the time changes of the process conditions and maintain the prediction accuracy. Usually, adaptive models are designed under the condition that the soft sensor only uses one adaptive model, even though it is difficult for a single adaptive model to maintain the prediction accuracy for all process conditions. Thus, previous studies have proposed strategies for selecting an appropriate adaptive model. However, this strategy has the following problems: limited types of adaptive models are available and the schemes are inefficient in terms of improving the prediction accuracy. To overcome these problems, a new strategy for adaptive models selection was developed in this study. The effectiveness of the proposed method was investigated by analysis of a numerical simulation data set and an actual process data set.