Rent-seeking experiments have been the subject of a great deal of research. In this paper, we focus on the repeated rent-seeking game by comparing fixed matching and random matching among opponents. Instead of using the winner-take-all form of the Tullock model, this paper's prize allocation is based on proportional sharing. I attempt to answer the following questions: What is the difference in participants' behavior between the two types of matching? Will participants overbid with respect to the Nash equilibrium? Will collusion behavior occur in either type of matching? The answers to these questions are of both practical and theoretical importance for rent-seeking analyses. Based on experiments with real money payments, 48 participants participated in two types of matching games: random matching (with random opponents) and partner matching (with fixed opponents). By analyzing the observed behavior of the participants, the results show that participants bid more than the Nash equilibrium in both types of matching; however, this result is more obvious in the random-matching group. Learning occurred in both types of matching. Overbidding in random matching is not reduced by learning. Participants in partner matching engage in collusive behavior, and they bid lower than the Nash equilibrium. The participants have low average bids and high profits. Conversely, participants in random matching have high average bids and low profits. We find that participants in more convergent groups have higher profits. The results also show that a female's profit is generally lower than a male's profit, and more males converge to their opponents than females. The regression result shows that neither participants' status as a postgraduate or undergraduate student nor whether they are majoring in economics or another subject has any influence on the bid behavior in our experiment.