Monsoon depressions form during the Southwest Indian Monsoon over the Bay of Bengal and provide copious rainfall over the eastern and central parts of the country. Since these depressions form over sea, a region of data scarcity, satellite data provides only source of information of the meteorological system. Furthermore, for short-range prediction, it is extremely important to have accurate initial conditions for better model performance. In this study, effects of three dimensional variational (3DVAR) assimilation of the Quick Scatterometer (QuikSCAT) data is used in the simulation of two monsoon depressions (MDs) that formed during 2-5 September and 27-30 September 2006 using the Weather Research and Forecast (WRF) modeling system. The National Center for Environmental Prediction - Global Forecast (NCEP-GFS) fields were used for the initial and lateral boundary conditions. Two model runs were employed in this study; first a control (CTRL) or a base run without any data assimilation and another a 3DVAR run in which QuikSCAT data was assimilated using the 3DVAR assimilation. The model results from both runs were compared with one another as well as with Tropical Rainfall Measurement Mission (TRMM) observations and Global Analysis (GFS-ANL) fields. The results of the time and area averaged vertical profile of relative vorticity over monsoon depressions indicate that the 3DVAR run is in closer agreement with GFS-ANL as compared to the CTRL run. Furthermore, the well-known temperature structure of a monsoon depression (cold core at low levels and warm core at upper levels) is better simulated by the 3DVAR run. While there is a clear and marked positive impact of ingesting the QuikSCAT data in terms of simulated precipitation for the depression that formed during 27-30 September 2006, improvement in the simulated rainfall due to QuikSCAT assimilation is slight for the other depression that formed during 2-5 September 2006. Consistent with the above observations, there is a clear improvement in the quantitative measures of the skill scores with lower bias, lower false alarms and higher probability of detection for almost all rainfall thresholds for the model runs which have assimilated the QuikSCAT observations.