Background Youth living with HIV with perinatal infection spend a lifetime taking antiretroviral treatment (ART) to suppress the virus, and face significant challenges to successfully maintaining ART adherence. Tools to measure adherence include self-report, medication event monitoring system (MEMS) pill bottle caps, pill counts, and plasma or hair drug levels; however, the inter-rater agreement between child and caregiver self-report has not been validated in an African setting. This study aims to assess inter-rater agreement between child and caregiver self-reports, compared to reporting from MEMS pill bottle caps. Methods This was a secondary analysis of a cluster-randomized trial to evaluate an intervention for children living with HIV, conducted at the Academic Model Providing Access to Healthcare in western Kenya. We analyzed data from 285 child-caregiver dyads to compare adherence self-reported by children and their caregivers, and subsequently compared all self-reports to adherence reported by MEMS pill bottle caps to determine whether child or caregiver self-reports aligned more closely with adherence measured by MEMS. Results Children and their caregivers reported similar levels of adherence and numbers of missed doses in the past month, and both reports were similarly associated with adherence reported by MEMS pill bottle caps. Children with a caregiver that was not a biological parent were significantly more likely to report more missed doses than their caregiver. The correlation coefficient for the relationship between the child and caregiver self-reports was 0.71; for the relationship between child report and MEMS was 0.23; and for the relationship between caregiver report and MEMS was 0.20. Both children and caregivers under-reported non-adherence compared to MEMS data. Conclusion Children and caregiver self-reports were generally similar in reporting adherence and were not highly correlated with MEMS reports of adherence, with children and caregivers reporting higher level of adherence than the MEMS data. This may indicate that children and caregiver reports are similarly inaccurate or biased; however, further research with larger sample sizes is required to further understand the differences in these reports.