This paper investigates the forecasting performance of different time-varying BVAR models for Iranian inflation. Forecast accuracy of a BVAR model with Litterman’s prior compared with a time-varying BVAR model (a version introduced by Doan et al., 1984); and a modified time-varying BVAR model, where the autoregressive coefficients are held constant and only the deterministic components are allowed to vary over time. Application using quarterly data of the Iranian economy from 1981:Q2 to 2006:Q1 shows that the performance of different specifications of time-varying BVAR models for forecasting inflation depends on the number of lags, hyper parameter that controls time variation, and forecast horizons. Our results, however, show that the modified time-varying BVAR model performs much better than other models regardless of the factors above.