An improved recursive Bayesian estimation approach with high efficiency and accuracy is proposed for the estimation of transformer tap positions. Transformers with suspect tap positions are identified firstly， then state estimation residuals for each tap position of the suspicious transformer are calculated. The posterior probability of all tap positions can be obtained with recursive Bayesian estimation approach. Furthermore， the computational efficiency are improved from the following aspects: a linear correction method and a prediction-correction method are proposed to estimate the residuals effectively; a heuristic method is given to reduce the set including the right tap position during the recursive process; and a stopping criterion for the recursive process is presented to avoid unnecessary estimation time. Simulations are carried on the IEEE 39-bus system to illustrate that the approach is reliable and with high efficiency， and thus has the good prospect of online applications.