Most studies on the perception-production link have assumed a synchronous relationship, according to which gains in perception transfer to production rapidly and efficiently. However, time-lagged and asymptotic relationships are also possible, in which case perception would guide production at a later stage or production would improve only once perception has reached a high level of accuracy. The present study investigates the temporal dynamics of the perception-production link by modeling English speakers’ ability to perceive and produce L2 Spanish stops over time. Mixed-effects modeling of the d’ perception and voice onset time (VOT) production data demonstrated significant development in both areas. Time-lagged change models indicate that a change in d’ was significantly related to decreasing VOT in L2 /p/ at the following session, but no significant relationships emerged between perception and production of L2 /b/. A range of patterns emerged at the individual level. Results are interpreted in light of contemporary models of L2 speech learning.