Introduction: Although prevention of mother-to-child HIV transmission (PMTCT) programs are widely implemented, many children do not benefit from them because of loss to follow-up (LTFU). We conducted a systematic review to determine the magnitude of infant/baby LTFU along the PMTCT cascade. Methods: Eligible publications reported infant LTFU outcomes from standard care PMTCT programs (not intervention studies) at any stage of the cascade. Literature searches were conducted in Medline, Embase, Web of Knowledge, CINAHL Plus, and Maternity and Infant Care. Extracted data included setting, methods of follow-up, PMTCT regimens, and proportion and timing of LTFU. For programs in sub-Saharan Africa, random-effects meta-analysis was done using Stata v10. Because of heterogeneity, predictive intervals (PrIs; approximate 95% confidence intervals of a future study based on extent of observed heterogeneity) were computed. Results: A total of 826 papers were identified; 25 publications were eligible. Studies were published from 2001 to 2012 and were mostly from sub-Saharan Africa (three were from India, one from UK and one from Ireland). There was extensive heterogeneity in findings. Eight studies reported on LTFU of pregnant HIV-positive women between antenatal care (ANC) registration and delivery, which ranged from 10.9 to 68.1%, pooled proportion 49.08% [95% confidence interval (CI) 39.6–60.9%], and PrI 22.0–100%. Fourteen studies reported LTFU of infants within 3 months of delivery, range 4.8–75%, pooled proportion 33.9% (27.6–41.5), and PrI 15.4–74.2. Children were also lost after HIV testing; this was reported in five studies, pooled estimate 45.5% (35.9–57.6), PrI 18.7–100%. Programs that actively tracked defaulters had better retention outcomes. Conclusion: There is unacceptable infant LTFU from PMTCT programs. Countries should incorporate defaulter-tracking as standard to improve retention.