In the modelling of non-Gaussian time series, one strategy is to retain the general autoregressive moving average framework and allow the white noise to be non-Gaussian. In this work, we are interested in correlated data exhibiting asymmetry by adopting a non-Gaussian autoregressive model with Azzalini's skew normal innovations. The moments and conditional maximum likelihood estimators of the parameters are derived, and their limit distributions are studied by Monte Carlo simulation. Finally, the flexibility of this model is illustrated by fitting it to a real time series.