Abstract In the dual-antenna GNSS/MEMS integration attitude determination, MEMS accelerometer, with its low precision and low cost, usually contains relatively large noise, which results in low accuracy of accelerometers derived horizontal angles and impacts the accuracy and reliability of attitude determination. The accelerometer signal has high-frequency vibration noise caused by the inherent structure of the vehicle. Therefore, instead of the traditional time-domain denoising methods, the Butterworth low pass filter with sliding window based on frequency domain method is designed to smooth accelerometer in this paper. A vehicular test result shows that the filter has a better effect on denoising and the accuracy of accelerometers derived horizontal angles is improved by about four times, reaching (0.643°, 0.546°). Additionally, the integration attitude solution is also improved from (0.124°, 0.738°, 0.532°) to (0.122°, 0.074°, 0.052°), where horizontal angles accuracy has an improvement about one order of magnitude. It is verified that the attitude solution after Butterworth low pass filtering is more accurate and reliable.