We design a novel sequential public goods experiment to study reciprocity, or conditional cooperation. In contrast to the standard simultaneous contribution game, our sequential design provides direct evidence on how subjects condition their own contributions on the contributions of other subjects in the experiment. We develop a simple but useful behavioral-type classification procedure and use it to analyze the data from this design. Our results inform two fundamental hypotheses: (1) subjects’ types are persistent over an experiment; and (2) the types of subjects included in a group affects a group’s ability to sustain cooperation. These hypotheses are often assumed in the public goods literature, yet neither has been directly supported. We find support for both hypotheses. Moreover, we provide a simple summary statistic that, we show, predicts group cooperative dynamics remarkably well.