In this study, a mesoscale model is used to investigate the possible reasons for the failure of the operation numerical models in the prediction of the record-breaking snowstorm of the 24-25 January 2000. The success of the high-resolution control simulation shows that the storm could have been well forecasted with conventional data in real-time. Various sensitivity experiments suggest that strong sensitivity to the model initial conditions and insufficient model grid resolution conditions may have contributed significantly to the difficulty of the real-time operational numerical forecast. Increased model grid resolution and improved initial analysis helps better forecast of the cyclone strength and location but the most dramatic improvements are seen in the precipitation forecast. The benefits of increased model grid resolution come from better representation of the moist processes and the associated diabatic heating feedback. The improvement in the reanalysis does not come from one (or a few) "key" sounding observations. However, omitting even single sounding from initial conditions produces significant changes in precipitation forecast, although mean sea-level pressure is influenced only slightly. This strong sensitivity of precipitation forecast to small initial differences is consistent with rapid error growth of differences at smaller scales and may have contributed to the difficulty of predicting accurately the record-breaking snowstorm. Even faster error growth at smaller scales is expected in higher-resolution simulations. Rapid error growth at smaller scales may well be one of the key factors in limiting the predictability of quantitative precipitation forecast for mesoscale.