Abstract: Acceleration is a key physical quantity in structural dynamic testing. The insufficient accuracy of conventional acceleration measurement systems in low-frequency signal components remains a challenge. To evaluate low-frequency acceleration measurement errors and perform compensation, shaking table tests were conducted using a high-precision strong-motion instrument as the reference. The results show that the measurement errors of conventional acceleration measurement systems are observed across the frequency range of 0.1-0.5 Hz, with the most pronounced errors occurring in the 0.1–0.3 Hz band. At 0.1 Hz, the relative peak error of IEPE accelerometers reaches up to 53%, while that of servo accelerometers is about 30%. At 0.3 Hz, the error of servo accelerometers decreases to within 5%, and that of IEPE accelerometers is reduced to about 10%. When the frequency increases to 0.5 Hz, the errors of all sensors decrease to relatively low levels. To address these errors, a nonparametric offline compensation method based on regularized least squares is proposed. By introducing transfer-function estimation based on power spectra and combining regularization with phase-centering-based time-domain constraints, the proposed method alleviates the instability and high-frequency noise amplification caused by conventional frequency-domain division. Under seismic excitation, after compensation, the relative peak errors of servo and IEPE accelerometers are reduced to within 1.5% and 4%, respectively, while the time-domain correlation coefficients are both improved to above 0.98, significantly enhancing the accuracy of low-frequency acceleration signals.