The starter motor noise is usually well identified by the customers since it is operated while the engine is quiet, and leaves distinct impressions of the vehicle. In this study the design of experiments(DOE) was applied to the identification of the noise sources of direct drive starter motor since this process usually requires lengthy analysis and elaborate experiments. In the first stage of the test, five controllable factors(alignment and dynamic unbalance of armature, tightening torques of T/bolt and center bracket bolt, and alignment of the center bracket-yoke-rear bracket), excluding static unbalance, are sorted out of all possible factors. Test results showed that the dynamic unbalance and misalignment of armature are the major factors. However, the error level of the first test was relatively high, indicating that there might be some missing major factors. In the second stage test the results showed that both static and dynamic unbalances are the dominant factors contributing to more than 80% of the overall noise, while the misalignment contributes around 12%. Error of the second test was about 4% that could be considered satisfactory. The noise level of the optimal product was predicted to be reduced by 19dBA, and verification test showed the average noise reduction of 16.8dBA with the standard deviation of 3.2dBA, and proved the usefulness of the whole DOE process.