We present a preliminary evaluation of the performance of three different cumulus parameterization schemes in the ICTP Regional Climate Model RegCM3 for two overlapping domains (termed "lbig" and "small") and horizontal resolutions (50 and 25 km) in the Caribbean area during the summer (July-August-September). The cumulus parameterizations were the Grell scheme with two closure assumptions (Arakawa-Schubert and Fritsch-Chappell) and the Anthes-Kuo scheme. An additional sensitivity test was performed by comparing two different flux parameterization schemes over the ocean (Zeng and BATS). There is a systematic underestimation of air temperature and precipitation when compared with analyzed data over the big domain area. Greater (∼ 2°C) and smaller (∼0.9°C) negative temperature biases are obtained in Grell-FC and Kuo convective scheme, respectively, and intermediate values are obtained in Grell-AS. The small domain simulation produces results substantially different, both for air temperature and precipitation. Temperature estimations are better for the small domain, while the precipitation estimations are better for the big domain. An additional experiment showed that by using BATS to calculate the ocean fluxes in the big domain instead of the Zeng scheme, precipitation increases by 25% and the share of convective precipitation rose from 18% to 45% of the total, which implies a better simulation of precipitation. These changes were attributed to an increase of near surface latent heating when using BATS over the ocean. The use of BATS also reduces the cold bias by about 0.3-4.0°C, associated with an increase of minimum temperature. The behavior of the precipitation diurnal cycle and its relation with sea breeze was investigated in the small domain experiments. Results showed that the Grell-Arakawa-Schubert closure describes better this circulation as compared with Grell-Fritsch-Chappell closure. © Springer-Verlag 2006.