Long-term historical air temperature records of four stations from Global Historical Climatology Network-Daily are analyzed in this study. By applying detrended fluctuation analysis of the second order to the monthly anomalies, different long-term correlations are found in different time periods at both the maximum and minimum temperatures, which indicate the existence of internal stochastic trend. By generating surrogate data with the same long-term correlations and data length, internal stochastic trends are estimated with confidence probability intervals of 95%and 99%provided. We find the longer data length, the shorter confidence probability interval we have; the stronger long-term correlation, the wider confidence probability interval is obtained. By comparing the temperature trends observed from the historical temperature records with the corresponding confidence probability intervals of the internal stochastic trends, significant external trends can be detected. We find that except for the maximum temperature in SAGINAW MBS INTL AP, temperatures from the four stations all show significant external trends when long historical data (> 100 years) are considered. However, if only the past 30 years are taken into account, the observed trends are still not strong enough to exceed the confidence probability interval. Although we cannot exclude the existence of external trends, considering the possible influence from internal stochastic trends, the external trends are not significant. From this detection method, we can judge, in the context of global warming, whether an observed trend is significantly induced by external forcing. Therefore, it is useful for our further study targeting the internal (external) climatic impact factors.