Let X i (i = 1, 2, . . .) be a sequence of subexponential positive independent and identically distributed random variables. In this paper we offer two alternative approaches to obtain higher-order expansions of the tail of n i=1 X i and subsequently for ruin probabilities in renewal risk models with claim sizes X i . In particular, these emphasize the importance of the term P( n i=1 X i > s, max(X 1 , . . . , X n ) ≤ s/2) for the accuracy of the resulting asymptotic expansion of P( n i=1 X i > s). Furthermore, we present a more rigorous approach to the often suggested technique of using approximations with shifted arguments. The cases of a Pareto-type, Weibull and Lognormal distribution for X i are discussed in more detail and numerical investigations of the increase in accuracy by including higher-order terms in the approximation of ruin probabilities for finite realistic ranges of s are given.