With the increasing interest of using inertial measurement units (IMU) in human biomechanics studies, methods dealing with inertial sensor measurement errors become more and more important. Pre-test calibration and in-test error compensation are commonly used to minimize the sensor errors and improve the accuracy of the walking speed estimation results. However, the performance of a given sensor error compensation method does not only depend on the accuracy of the calibration or the sensor error evaluation, but also strongly relies on the selected sensor error model. The best performance could be achieved only when the essential components of sensor errors are included and compensated. Two new sensor error models, with the concerns about sensor acceleration measurement biases and sensor attachment misalignment, have been developed. The performance of these two error models were evaluated in the shank-mounted IMU-based walking speed/inclination estimation algorithm with a comparison of an existing error model. The treadmill walking experiment, conducted at both level and incline conditions, demonstrated the importance of sensor error model selection on the spatio-temporal gait parameter estimation performance. Accurate walking inclination estimation was made possible with a newly developed sensor error model.