Hidden Markov models (HMMs) are statistical models widely used for studying dynamic processes that cannot be observed directly or are governed by a hidden layer. This chapter introduces the framework of multichain HMMs (MHMMs), which provides a unified view to existing models in the literature. It demonstrates the utility of these models in ecological dynamics modeling. The chapter discusses inference for MHMMs in the context of the expectation-maximization (EM) algorithm. It explains why exact inference remains tractable for some structures of MHMMs. A promising way of addressing both selection of the number of states and the conditional independence relationships in MHMMs is Bayesian statistical modeling. This offers the possibilities, first, to handle non-bounded a priori numbers of parameters using Dirichlet processes and, second, to replace the separate estimation of several models with the estimation of posterior distributions of random variables in nested models, using spike-and-slab priors.