Ozone in the troposphere is a harmful gas for life on Earth, so its study and control is very important for us. In this paper, the ozone concentration data of Mexico City from 1992 to 2015 are analyzed as annual time series using hidden Markov models (HMM). It is considered that the different ozone generators form a sequence of states in a stochastic process that can be modeled with a Markov chain, and the ozone concentration measurements from the monitoring centers are random variables whose distribution function depends on the state of the Markov chain at time t. Two distribution functions were considered for the observed data, the normal and the Gamma, and the model parameters were estimated using EM method. One, two and up to seven states for the Markov chain were carried out. Finally, 222 models were estimated and the best model was selected for each year using the Bayesian information criterion, BIC. It is concluded that this model describes the data well.