A heavy snow storm paralyzed Washington, D. C. on 25 January 2000, with a 24-h maximum snowfall of approximately 16 inches. The storm began on 23 January as a surface low located in the southeastern United States near the coast of the Gulf of Mexico, and then it moved northward along the east coast. Most operational models failed to predict the heavy snow that fell over Washington, D. C.. The control simulation using the Penn State/NCAR mesoscale model version 5 (MM5) and conventional data analysis procedures, also failed to correctly simulate the location and the intensity of the snow storm. The simulated storm is too weak and the T-bone shape of the snow distribution shown in satellite imagery is not captured on the other hand. Observations from the NASA Total Ozone Mapping Spectrometer (TOMS) ozone data on 24 January show a strong ozone anomaly located near the coast of the Gulf of Mexico. This feature was missing in the original MM5 analysis. In order to determine the extent of the forecast failure caused by the uncertainties in model initial condition, the TOMS ozone data on 24 January are incorporated into the initial condition. A four-dimensional variational assimilation (4D-Var) is carried out using a statistical correlation model between the total ozone and vertically integrated Ertel potential vorticity (IPV). This correlation model is developed using a linear regression method assuming a simple relationship between the vertically integrated ozone and IPV: Ozone=A * (IPV) + B, where A and B are determined statistically based on 20 mesoscale forecasts at 30-km resolution (with hourly output) and the TOMS ozone observations in the 10-day period from 15-25 January 2000 over the United States. The linear relation between IPV and ozone is found to depend on the latitude, with a strong linear correlation in middle latitudes and a weak linear correlation in the tropics and high latitudes. Preliminary numerical results of TOMS ozone data assimilation and forecast, and plans for an improved use of TOMS ozone data for mesoscale prediction will be presented at the conference. Other possible reasons for the forecast failure are also studied.