Most empirical studies on the returns to education use current income to proxy for lifetime income due to the lack of longitudinal data. This simplification is found to cause biased estimates and the result is conditional on being employed. This paper quantifies the returns to education with heterogeneity in employment rates, career lengths, and income growth rates. Using data from China, this paper attempts to account for these differences across the life-cycle and estimates the returns to education in terms of lifetime income when actual lifetime earnings data are not available. The model clarifies the mathematical relationship between conditional current returns to education, unconditional current returns to education, and unconditional lifetime returns to education. This new approach explains how employment rates, career lengths, and income growth rates affect the direction and magnitude of the bias in estimating the returns to education.