Abstract: Facing the intelligent detection target of pavement roughness, this paper took the widely used smart phone as the vibration detection method of driving vehicle, and carried out a series of driving experiments under different vehicles, different speeds and other conditions. The average absolute deviation of the vertical vibration acceleration produced in the driving process was used as the driving vibration acceleration index (VAI), the relationship between the driving vibration acceleration, the driving speed and the pavement roughness was analyzed, and a parameter called speed sensitivity (SS) was proposed. Results show that the VAI has a significant linear correlation with the driving speed when the pavement roughness is almost the same, which can be characterized by the parameter called speed sensitivity, and the pavement roughness has also a significant linear correlation with the speed sensitivity parameter. When the international roughness index (IRI) is larger than 2 mm/m, the correlation becomes more sensitive. Based on the above analysis, the IRI detection model called IRI-SS model was put forward. The results of preliminary driving test show that the average detection accuracy of IRI by using this method is more than 85% in total, and the highest average detection accuracy reaches up to 93%, which shows the possibility of practical application. This detection method provides a new possible way for the intelligent detection of pavement roughness.