To determine the acceleration and deceleration rate for the operating speed prediction model, the continuously driving tests of passenger car, bus, and heavy trucks were carried out on two-lane roads in a mountainous area. Longitudinal acceleration of vehicles was collected, and the cumulative frequency curves of peak acceleration and deceleration rate were analyzed. The statistical distribution and the eigenvalues of acceleration (a_x) and deceleration (a_b) were obtained, and the models of a_x and a_b with road geometry parameters as independent variables were established. The results show that: due to that the magnitude of a_b of passenger cars is about twice higher than a_x, the a_x and a_b could not be simplified to a same fixed value. Slope breakpoint of cumulative frequency curve of a_x and a_b does not appears on the 85^th percentile but near the 95^th percentile. The a_x and a_b of large buses display very close amplitude, and a_b of heavy-duty trucks is significantly greater than a_x. a_x and a_b of heavy-duty trucks are lower than large buses'. There is a negative correlation between longitudinal acceleration of passenger car and curve radius, and a positive correlation with curve deflection angle.