Due to fast response and adjustable damping force with the application of magnetic fields, magnetorheological suspension of all-terrain vehicles (ATV) has significant advantages in vibration suppression, especially for the complicated driving conditions. However, it is a challenge to identify the vehicle driving conditions in the case of noise or abnormal sensors. This paper focused on a fusion technology of multi-sensor information eigenvalues based on D-S (Dempster-Shafer) evidence theory to improve the accuracy of driving cycle identification. Firstly, the improved distance estimation method was used to select and identify the sensor eigenvalues related to driving conditions, and then the noise and outliers of sensors were treated as uncertain information by interval estimation. The identification results of feature layer were fused by D-S synthesis, and the driving condition identification of ATV was completed based on the decision rule of feasible interval. Finally, the validity of the decision level fusion method with D-S evidence theory was verified in Carsim simulation software.