The performance of cumulus parameterization schemes in the short range prediction of rainfall during onset phase of the Indian Southwest Monsoon (ISM) was evaluated using Fifth-Generation Pennsylvania State University / National Center for Atmospheric Research Mesoscale Model (MM5). MM5 model was used to predict rainfall at 30 km resolution up to 72 h over the Indian subcontinent for each day during the period 1–30 June 2002, which corresponds to the onset phase of the ISM. Experiments were performed with 5 different cumulus parameterization schemes of Anthes–Kuo (AK), Grell (GR), Betts–Miller (BM), Kain–Fritsch (KF) and Kain–Fritsch2 (KF2). Rainfall prediction assessment was made over five zones through comparison with corresponding APHRODITE gridded precipitation data and for selected location with station observations by analyzing the statistical parameters of correlation coefficient, mean absolute error and Hanssen–Kuipers score. Monthly mean zone-wise rainfall was well predicted by GR and AK schemes up to 48 hours and slight overestimation beyond. GR scheme predicted higher rainfall over west coast, central parts of India and low rainfall over southeast peninsula. BM and KF schemes showed overestimation with prediction of rainfall over dry southeast peninsula. All the schemes underestimated the coefficient of variability (CV) over all the five zones. AK and GR schemes had the mean and CV nearer to the APHRODITE observations, with AK scheme slightly better than GR scheme over Zones 1, 2 and 3 while GR scheme had the best agreement over Zones 4 and 5. GR scheme had also shown higher CC values and lower MAE over most of the zones up to 72 h, while BM had the least predictability with lower CC and HK scores and higher MAE over most of the zones. Over Western Ghats, the uncertainty limits could be higher than shown due to dominant heavy rains. Of the ten stations selected for verification, GR scheme had shown better prediction with significant positive CC values at nine of the ten stations and consistently lower MAE values and higher HK scores. Further analysis has shown that GR scheme predicted higher grid-scale and nighttime rainfall agreeing with earlier studies concerning monsoon rainfall. All other schemes predicted the features contrarily with higher convective and daytime rainfall. GR scheme alone was found to have provided the best prediction considering the mean monthly, daily zone-wise and station rainfall predictions. The present study concludes that GR cumulus parameterization scheme is the most suitable at 30 km resolution.