A model fusion method by weighted average is proposed; weights are optimized through cross-validation or 10-fold cross-validation. This method produces a combined model which incorporates models under similar conditions into data in hand. A model to predict rice yield in Fukushima area is improved by a model for Ibaraki area in terms of predictability. Furthermore, a model for discrimination of crab by sizes and sex exemplifies a similar effect when models in different forms are integrated.