This research on the asymmetric information problem in the automobile comprehensive insurance of Taiwan emphasizes on separating it into asymmetric information about risk type and asymmertic information about the risk aversion attitude. Besides on the point about the heterogeneity between new car and old car policies, we also analyze pair-wisely on the policies of type A, B and C. By conditional correlation analysis, we find quite different evidences between them: there are asymmetric information about risk type between type A and C as well as type B and C, and there is cherry picking story between type A and B. Our results are robust to the car age. The only difference is the threshold effect about the claim amount which comes from that new car get stronger asymmetric information. In the full sample, the audit ftom insurance company is more severe when the claim amount increase. In sub-sample of old car, the insured becomes more risk averse when the loss amount increases. Our conclusion also won't change by the choice of tested data area. The only difference is in the strength of their asymmetric information.