The accuracy of temperature probe calibration in a constant temperature bath is governed by the stability and uniformity of the bath. In the available literature, there have been efforts in designing baths with better stability and uniformity. This work is distinguished from the previous efforts by presenting a novel data processing method that improves the calibration performance without the need to improve the inherent qualities of the bath. The stability limitation is addressed by filtering out the time dependent components of temperature readings, and then the resulting calibration performance becomes limited only by the uniformity. This work provides the theory and guidelines for filtering the time dependent signal components such as sampling rate and duration based on the stability of the bath. The method is experimentally validated by applying it to a baseline bath and the results are compared to those from a high-end bath. As a figure of merit, the probability of obtaining a calibration value within ±0.75 mK of the high-end bath is reported. In a calibration scenario in which the uncertainty is predominantly governed by the stability, only 10% of the calibration values are within the aforementioned limits with the conventional method. With the application of the novel method, all of the calibration data meets the criterion.