Kawakami et al. (2008) offer an approach training intervention to strengthen women’s implicit identification with math. This intervention is especially interesting regarding data suggesting that the low implicit identification with math among women and the gender gap in science reinforce each other (Nosek et al., 2009). Nevertheless, Kawakami et al.’s data only provide quite modest evidential value in favor of the effectiveness of this intervention (notably because of the two critical p values being very close to 0.05 and of the small sample sizes). In the present manuscript, we offer a preregistered replication of Kawakami et al. with a substantially larger sample size and a novel implementation of approach and avoidance (the VAAST; Rougier et al., 2018). In a Pilot Experiment (N = 150), we validate the VAAST-based approach/avoidance training as a way to create identification for novel stimuli ($d_s$ = 1.17, p < .001). We then replicate Kawakami et al.’s work, revealing that women who approached math (instead of avoiding it) had a higher identification with math ($d_s$ = 0.30, p = .037). This preregistered replication increased evidential value in favor of the effectiveness of the approach training by a factor of 2.57, now providing “strong” support for the effectiveness of the training (Jeffreys, 1961). A meta-analysis of the original data and the replication revealed a small-to-medium effect of this intervention ($d_s$ = 0.40, $CI_(95%)$ [0.14; 0.65]). These results are discussed in regard to theories explaining how actions affect evaluative response as well as current interventions in the literature.