The surface temperature outcomes of determined forecasts of the European Centre for Medium range Weather Forecasts (ECMWF) during the period from January 2007 to December 2010 are examined with root mean square error (RMSE) and the error is corrected by utilizing the methods of unitary linear regression, multiple linear regression, unitary bias removed with unitary lead time forecast and bias removed mean with multi lead time forecasts, respectively. The results show that all of the four methods could considerably reduce the ECMWF forecast errors for multi lead time forecasts, generally about 1℃. The forecast skill of the unitary linear regression is higher than that of two consensus forecast methods considering multi lead time forecast outcomes for short range forecast. While two kinds of consensus forecasts have higher and more stable forecast skills for medium range forecast. The correction methods considering multi lead time forecast outcomes could reduce the forecast error more stably especially when the error is large.