Three numerical weather forecast models, MM5,COAMPS and WRF, operating with a joint effort of NOAA HU-NCAS and Jackson State University (JSU) during summer 2003 and winter 2003-2004 have been chosen to study their forecast skills against observations. Grid resolution of 27 km for coarse domain nested with high resolution domain of 9 km were chosen to represent the current mesoscale model. The forecasts with the length of 36h were performed to output the results with 12h interval. The key parameters used to evaluate the forecast skill include 12h accumulated precipitation, sea level pressure, surface wind speed and surface temperature. Precipitation was evaluated statistically using conventional forecast skill scores, Threat Score (TS) and Bias Scores (BS), for different threshold values based on 12h rainfall whereas other variables were studied using statistical methods such as Mean Error (ME), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).