Kona Lows are subtropical cyclones that occur during the cool season in the north-central Pacific, producing a wide range of weather hazards. Operational NWP models perform poorly in forecasting kona-low development and track. In this paper we investigate the numerical simulation problem through a series of sensitivity studies using the PSU/NCAR mesoscale model MM5 and the Local Analysis and Prediction System (LAPS), with a focus on unraveling the role of convective latent heating. The distribution and impact of latent heating on the structure and evolution of the simulated kona low are diagnosed as a function of initial conditions, model grid resolution and the physical parameterizations used. The case choosen for this study is a strong kona low that occurred in February 1997, which has been the subject of previous observational analysis. Initially, MM5 is run in standard mode over a single 27 km resolution domain with Global Forecasting System (GFS) analysis and forecasts providing initial and boundary conditions. Then MM5 is run with three two-way nested domains (27, 9 and 3 km), with the innermost of domain moving with the kona low. Finally, enhanced initial conditions at 27 km resolution are produced by LAPS through assimilation of GOES9 satellite data and derived satellite products. LAPS allows MM5 to be diabatically initialized (hot start mode). The results show that an accurate simulation of central pressure minimum and track of the kona low is achieved when the MM5 is run with nested grids in hot start mode. The impact of latent heating on the structure and evolution of the kona low are investigated through application of quasi-geostrophic and potential-vorticity diagnostics.