Based on the air temperature prediction data from the TL799L91 model of the European Centre for Medium-Range Weather Forecasts (ECMWF), the prediction capabilities of time-lagged ensemble prediction and the traditional deterministic prediction for a single point were studied. The comparison between the time series of the observed data at Shenzhen and those from the deterministic prediction/time-lagged ensemble prediction is performed, which shows that: (1) The prediction capability of TL799L91 model can reach as long as 240 h for the air temperature of Shenzhen. Generally speaking, for deterministic prediction, the prediction error increases with the increase of forecast lead time, however, the latest prediction is not necessarily the best one. (2) The time-lagged ensemble prediction does help improve the prediction quality. The accuracy of the time-lagged ensemble prediction is directly related to the number of ensemble members. Generally speaking, the increase of ensemble members will lead to more accurate prediction results.The study of this paper shows that the time-lagged ensemble prediction is an effective way to make use of the numerical prediction of earlier initiation.