Automated surveys, by interactive voice response (IVR) or email, are increasingly used for clinical research. Although convenient and inexpensive, they have uncertain validity. We sought to assess the accuracy of longitudinally-collected automated survey responses compared to medical records. Using data collected from a well-characterized, prospective birth cohort over the first year of life, we examined concordance between guardians' reports of their infants' health care visits ascertained by weekly automated survey (IVR or email) and those identified by medical chart review. Among 180 survey-visit pairs, concordance was 51%, with no change as number of visits per baby increased. Accuracy of recall was higher by email compared to IVR (61 vs. 43%; adjusted OR = 2.5 95% CI: 1.3-4.8), did not vary by health care encounter type (hospitalization: 50%, ER: 64%, urgent care: 44%, primary care: 52%; p = 0.75), but was higher for fever (77%, adjusted OR = 5.1 95%CI: 1.5-17.7) and respiratory illness (58%, adjusted OR = 2.9 95%CI: 1.5-5.8) than for other diagnoses. For the 75 mothers in these encounters, 69% recalled at least one visit; among 41 mothers with two or more visits, 85% recalled at least one visit. Predictors of accurate reporting by mothers after adjusting for illness in the baby included increased age and increased years of education (age per year, β = 0.05, p = 0.03; education per year, β = 0.08, p = 0.04). Additional strategies beyond use of automated surveys are needed to ascertain accurate health care utilization in longitudinal cohort studies, particularly in healthy populations with little motivation for accurate reporting.