The aim of this paper is to demonstrate the relative performance of combining forecasts on inflation in the case of Tunisia. For that, we use a large number of econometric models to forecast short-run inflation. Specifically, we use univariate models as Random Walk, SARIMA, a Time Varying Parameter model and a suite of multivariate autoregressive models as Bayesian VAR and Dynamic Factor models. Results of forecasting suggest that models which incorporate more economic information outperform the benchmark random walk for the first two quarters ahead. Furthermore, we combine our forecasts by means and the finding results reveal that the forecast combination leads to a reduction in forecast error compared to individual models.