Five WRF configurations using different soil model, microphysics and planetary boundary layer parameterizations are compared with sounding data launched during a field campaign at APEX (Atacama Pathfinder EXperiment) site. The WRF model does a very good job forecasting PWV and temperature, wind speed and direction vertical profiles over the APEX site. Changes in microphysics parameterizations do not produce appreciable changes in humidity profiles. The Noah land surface model greatly improves the forecasts compared to the 5-layer thermal diffusion scheme. The analysis of daily synoptic conditions shows that difficulties in predicting the diurnal variation of wind direction in clear conditions and the occurrence of dry shallow layers in the atmosphere are some of the error sources in forecasts.