This article reports the investigation results of long-term stability tests on on-wafer measurement systems in frequency ranges up to 325 GHz conducted by the NMIJ-AIST. The radio frequency (RF) probe coordinates in the $X$ -, $Y$ -, and $Z$ -directions, and the tilt angle is automatically adjusted by performing RF signal detection. Calibrations and verification processes were repeated for over a year. The novelty of the study is that it optimizes measurement procedures for the test, demonstrates a long-term stability test, and compares the evaluated long-term stability to the effects of probe model, absorber, and calibration method. The standard deviation of the measured reflection coefficient ranged between −32 to −70 dB in magnitude, and 0°–10° in phase. The standard deviation of the measured transmission coefficient ranged from −43 to −80 dB and from 0.2°–2.7° in phase. The refined procedures contributed to improve the measurement reproducibility compared to the previous report. Furthermore, the effects of the chuck material and the calibration method were larger than long-term stability.