Most studies on public goods game reported that in a finitely repeated public good game subjects' contributions begin with an average contribution of about 50% of their initial endowment and decay toward the free riding level as the game progresses. Unconditional free ridings were seldom observed. To explain what causes this behavioral patterns, three hypotheses have been suggested; learning hypothesis, strategy hypothesis and reciprocity hypothesis. This paper investigates these hypotheses, and our findings from public goods experiments suggest that subjects' reciprocity is one of the main causes that produce subjects' behavioral patterns.