Purpose The noise measurement on magnetoresistive (MR) sensors is generally conducted by techniques including single-channel data sampling and fast Fourier transform (FFT) analysis as well as two-channel cross-correlation. The single-channel method is easy to implement and is widely used in the noise measurement on MR sensors, whereas the two-channel method can only eliminate part of the system noise. This study aims to address two key issues affecting measurement accuracy: calibration of the measurement system and the elimination of system noise. Design/methodology/approach The system is calibrated by using a low-noise metal film resistor in that the system noise is eliminated through power spectrum subtraction. Noise measurement and analysis are conducted for both thermal noise and detectivity of magnetic tunnel junction (MTJ) sensor. Findings The thermal noise measurement error is less than 2%. The detectivity of the MTJ sensor reaches 27 pT/Hz1/2 at 2 kHz. Originality/value This study provides a more practical solution for noise measurement and system calibration on MR sensors with a bias voltage and magnetic field.