We study the implementation problem when players are prone to make mistakes. To capture the idea of mistakes, Logit Quantal Response Equilibrium (LQRE) is used, and we consider a case in which players are almost rational, i.e., the sophistication level of players approaches infinity. We show that quasimonotonicity, a small variation of Maskin monotonicity, and no worst alternative are the necessary conditions for restricted Limit LQRE (LLQRE) implementation. Moreover, these conditions are sufficient for both restricted and unrestricted LLQRE implementations if there are at least three players and if at least one player has a state-independent worst alternative.