Based on the existing algorithms, an algorithm of vehicle speeds for anti-lock braking system (ABS) was investigated using parameter adaptive Kalman filter and ABS control inputs. With the four wheel speeds as inputs, the vehicle braking deceleration was estimated utilizing ABS control inputs. Then the speeds on both left and right sides were estimated independently using information of wheel speeds and deceleration. According to different slip conditions, the covariances were tuned using the parameter adaptive Kalman filter. At last, the proposed algorithm was evaluated using real test data of ABS braking, and results show that the estimated speeds are reasonable under various road conditions.