In this paper, two-phase reverse C-K-M and M-K-C neural mapping models are developed for modeling dynamically changing non-stationary process. Complementary two mapping structure are designed. The first mapping model, C-K-M neural network interpreted as a coarse model, approximates underlying process while the second M-K-C neural network is designed for obtaining fine model. Using the presented two-phase map-ping scheme, adaptive segmentation procedure is also developed to detect model change time. Focuses are given to perform model identification and model change detection presenting new efficient two-phase adaptive detection scheme. Experimental results are given to verify the proposed approach could be useful for modeling dynamically changing non-stationary process with small sample size.