Dose evaluation by electron spin resonance (ESR) is usually accomplished by constructing a dose-response curve by measuring the peak-to-peak intensity of the dosimetric signals in the g = 2 region. In several cases, this signal is overlapped with others that can interfere with dose reconstruction. In this work a new method to correct the spectrum before the measurement of the signal intensity is proposed. Examples of dose determination of accumulated dose (AD) of two fossil teeth from southern Brazilian megafauna are given. One of them presents a dose-independent signal in the region of interest, and the validity of this method is shown. For the other, without interfering signals, no difference in the AD was found. This method can also be applied to retrospective dosimetry by ESR for any sample with dose-independent interfering signals, thus improving the accuracy in AD determination by ESR.