Abstract: The TAP(terrain aided position) likelihood function reflects the probability of the position of AUV (autonomous underwater vehicle) in space. Due to the strong nonlinearity and randomness of the terrain and the non-Gauss distribution of the measurement error, the likelihood function also shows the characteristics of non-Gauss. The error of TAP is closely related to the local topographic and measurement error. Because the existing method does not consider the local topographic features, the statistical confidence interval of the measurement error is only established, so the estimation results of the TAP are obviously smaller. In this paper, a jump model of the TAP position X^p is established. It can jump to any point in the searching interval, and the probability jumping to any point is positively correlated with the likelihood function of the point. When the confidence of the jump to a certain point is less than α, X^p will not jump to this point and this point is called the boundary point of the confidence interval. Assumed the sum of squares of the matched residuals in the confidence interval of the TAP is quadric surface, X^p is regarded as the parameter of the quadric surface, and the confidence interval of TAP with confidence 1-α can be obtained by the confidence interval estimation method of the surface parameters. The confidence interval obtained by new estimate method is larger than the existing method. The experimental results show that the confidence interval estimation will be abnormal when the measuring beam is less. The increase of the measurement beam can improve the estimation accuracy of the tidal and measurement errors, thus can promote the estimation accuracy of the confidence interval, but the compensation method under the condition of non-Gauss distribution of measurement error is still needed in further work.