Obesity is associated with osteoarthritis (OA), but few studies have used fetal origin to explore the association. Our study aims to disentangle the causality between birth weight, childhood obesity, and adult OA using Mendelian randomization (MR). We identified single nucleotide polymorphisms (SNPs) related to birth weight (n = 298,142) and childhood obesity (n = 24,160) from two genome-wide association studies contributed by the Early Growth Genetics Consortium. Summary statistics of OA and its phenotypes (knee, hip, spine, hand, thumb, and finger OA) from the Genetics of Osteoarthritis Consortium (n = 826,690) were used to estimate the effects of SNPs on OA. Multivariable MR (MVMR) was conducted to investigate the independent effects of exposures. It turned out that genetically predicted standard deviation increase in birth weight was not associated with OA. In contrast, there was a marginally positive effect of childhood obesity on total [odds ratio (OR) = 1.07, 95% confidence interval (CI) = 1.00, 1.15 using IVW], knee (OR = 1.13, 95% CI = 1.05, 1.22 using weighted median), hip (OR = 1.13, 95% CI = 1.04, 1.24 using IVW), and spine OA (OR = 1.12, 95% CI = 1.03, 1.22 using IVW), but not hand, thumb, or finger OA. MVMR indicated a potential adulthood body mass index-dependent causal pathway between childhood obesity and OA. In conclusion, no association of birth weight with OA was suggested. Childhood obesity, however, showed a causality with OA in weight-bearing joints, which seems to be a general association of obesity with OA.