Based on conventional surface and upper air observation data, the synopticscale circulation of the maximum 2 m temperature forecast failure in Shanghai during 17-18 March 2015 is analyzed. Using the realtime observations and numerical weather predictions, the causes of the forecast failure are investigated. The results show that the failure of the cloud cover forecast on 17 March was the main cause for the forecast failure. Meanwhile, the enhanced southeast winds further enlarged the forecast error. On 18 March, the misforecasted occurrence time of cold air advection into Shanghai mainly accounted for the forecast failure. The problems and recommendations in the forecast process were addressed, with the emphasis on the discrepancies between the realtime observations and the forecasts, forecast jumpiness and forecast uncertainty. Forecasters should have paid more attention to the realtime local and upstream observations, and different leadtime forecasts from different models. More important, operational ensemble forecasts in a probabilistic sense were highly recommended instead of traditional deterministic forecasts of a single model. Therefore, it is necessary to systematically assess the ensemble forecast performance, develop new application and interpretation of NWP products, and provide more products of multimodel ensemble forecasts with uncertainty information in public weather services.