When an explicit expression for a probability distribution function $F(x)$ can not be found, asymptotic properties of the tail probability function $\bar{F}(x)=1-F(x)$ are very valuable, since they provide approximations or bounds for system performance, and approaches for computing probability distribution. In this paper, we study tail asymptotic properties for the number of the customers in the $M/G/1$ retrial queueing system. For queueing systems, studies on asymptotic tails have mainly concentrated on the first order asymptotic behavior. To best our knowledge, there is no second order tail asymptotic analysis for retrial queueing models with heavy-tailed service time. Second order asymptotic expansions provide the refined asymptotic results from the first order approximation, and are often more difficult to obtain as expected. The main contribution of this paper is the second order asymptotic analysis for the $M/G/1$ retrial queue.