Background: Blood DNA methylation patterns are highly predictive of prenatal exposure to smoking and differential methylation has been associated with maternal alcohol use. We extended this to determine whether DNA methylation patterns in cord blood are associated with prenatal exposure to opioid, cannabis, and polysubstance use. We also evaluated whether DNA methylation patterns have predictive utility. Methods: We examined 932 mother-child pairs in the Boston Birth Cohort between 1998-2020 with cord blood DNA methylation and maternal substance use data. For each substance, we performed adjusted linear regression analysis at 865,859 CpG sites to identify related methylation changes. We generated scores using summary statistics for each exposure and assessed predictive ability using cross-validation and receiver operating characteristic curves. Specificity of methylation associations was evaluated by assessing overlap across exposure summary statistics, and using logistic regression for methylation scores, adjusted for concurrent use. Results: We identified methylation changes at 72, 21, and 1 novel CpG associated with prenatal exposure to opioids, cannabis, and polysubstance use respectively, at epigenome-wide significance (P<1e-6). Comparing CpGs across exposure groups and with previous consortia-generated summary statistics for smoking and alcohol revealed few overlapping CpGs across exposures. Methylation scores were highly predictive and exposure-specific, with area under the curve accuracy of 91% for opioids, 90% for cannabis, and 93%-98% depending on polysubstance number. Conclusion: We identified novel DNA methylation differences in cord blood associated with prenatal exposure and showed that methylation scores are highly predictive of substance exposures. These CpGs provide biologic insights for reducing the impact of substance exposure. These findings may serve as a biomarker of prenatal substance exposure for future studies and potential clinical utility.