The order of autoregression in a time series model refers to the number of lagged observations included as predictors. It determines the complexity and flexibility of the model in capturing temporal dependencies and patterns in the data.