Based on the daily surface air temperature (SAT) gauge data and global reanalysis datasets from 1961 to 2016, this study revealed the interannual variability of the SAT over Northwest China (NWC) in May and September via observational diagnosis and numerical simulations, and constructed their corresponding seasonal prediction models. The results are as follows: (1) The first modes of the SAT over NWC during May and September are characterized by a similar homogenous spatial pattern but with different interannual variations. (2) The positive anomaly of the SAT over NWC in May is related to the tropical zonal tripole anomalous convection (precipitation), corresponding to the tropical sea surface temperature anomaly (SSTA) during the decaying phase of La Niña. The extratropical teleconnection wave train excited by tropical convection anomalies leads to a barotropic anticyclonic (high pressure) anomaly over NWC, which increases the downward solar shortwave radiation and causes an increased local SAT. In September, the tropical zonal dipole convection (precipitation) anomaly associated with the tropical SSTA during the developing phase of La Niña can trigger the barotropic anticyclonic (high pressure) anomaly on the east and west sides of NWC, leading to a positive SAT anomaly over NWC. (3) Based on the SSTA predictors associated with the decaying and developing phase of La Niña, the seasonal prediction model for the SAT over NWC in May and September was established. The prediction skill in terms of the correlation coefficient during the independent prediction period (2006–2016) can reach 0.74 (0.62) in May (September), providing a reference for seasonal prediction of the SAT over NWC.