<p>Tilt angle measured by only one dual-axis accelerometer is always influenced by some noise from measurement and environment. Through combining several low-cost sensors, we can achieve a high accuracy sensor via reducing the noise. This paper presents an approach to enhance the accuracy of inclinometer based on four MEMS dual-axis accelerometers fusion. For any of the accelerometers, the tangent and cotangent function instead of traditional arcsine function are alternately chosen to solve for tilt angle for the purpose that the sensitivity of tilt angle keeps maximum constant value in full measurement range. The method of inter-quartile range is used to eliminate spurious data during signal acquisition. The modified Bayesian fusion with pre- and post-filtering are employed to reduce noise. Experimental results show that the error range and variance of the fused tilt sensor are reduced by an average factor of 2.53 and 4.45 compared to four individual tilt sensors.</p>