Abstract: This paper proposes hidden Markov models with Yeo-Johnson transformation for analyzing heterogeneous longitudinal data. The model assumes data from different hidden states to reveal the inner structure and transition mechanisms. Moreover, the model introduces the Yeo-Johnson transformation to handle non-normal observations. This paper employs Bayesian methods, including Markov chain Monte Carlo and Watanabe-Akaike Information Criterion, to obtain the estimates of unknown parameters and the number of hidden states. Simulation results show that the proposed methods can provide accurate estimates of the unknown parameters and the number of hidden states. Finally, the National Longitudinal Survey of Youth 1997 data set is analyzed using the proposed model and methods.