With time delay under consideration, temperature correlation matrixes are constructed based on the reanalysis of temperature data provided by National Centers for Environmental Prediction/National Center for Atmospheric Research and European Centre for Medium-Range Weather Forecasts. Results indicate that the correlation of global temperature decreases with lag time, and the rate is dependent on time lag. We divide the lag time (1—30 d) into three segments, i.e., 1—7 d, 8—20 d and 21—30 d according to the decrease rate of global average correlation coefficient Cglb. When the lag time is in a specific interval (8—20 d), Cglb is unstable, which may explain the difficulty in long range weather forecast of 10—30 d. The spatial distribution of the global temperature correlation keeps stable for different lag times, while the numerical change shows zonal distribution on the whole, and that the most of Asia and the equatorial central and eastern Pacific show countertrend to other parts of similar latitudes.