Introduction: Neurodevelopment delay is an important public health problem, affecting up to 15% of children. We aim to identify environmental risk factors that predict delayed trajectories of neurodevelopment using latent growth mixture modeling (LGMM).Methods: We use the Programming Research in Obesity, Growth, Environment and Social Stress, (PROGRESS) a longitudinal birth cohort in Mexico with repeated measures of Bayley Scales of Infant and Toddler Development (BSID) (n = 520). The BSID subscales for motor, cognition and language are assessed at four time points from 6 months to 24 months of age. Covariates include maternal hemoglobin, maternal IQ, socioeconomic status, maternal education level, child birthweight and child gender. We build parallel process LGMM to identify distinct subgroups (classes) with similar growth trajectories. We simultaneously model all three BSID subscales while adjusting for covariates, and allow for potentially nonlinear trajectories using quadratic growth factors. We then test for the association between blood metal concentrations (arsenic, cadmium, cobalt, chromium, cesium, copper, manganese, lead, antimony, selenium, zinc) collected during pregnancy and latent class membership.Results: We find that a four-class model best fits the data, and each class is characterized by distinct growth trajectories for the three BSID subscales. We interpret the classes as: high normal (4.2%), early normal (52.3%), late normal (35.2%), and early delayed (8.3%). The early delayed group has significantly lower exposures to zinc (p=0.05) at the second trimester of pregnancy, as compared with all other classes. Furthermore, the early delayed group has significantly lower concentrations of manganese (p=0.02), zinc (p=0.004) and copper (p=0.02) in child’s blood at birth as compared with all other classes.Conclusions: Latent classes of neurodevelopment may be differentially vulnerable to metal exposures.