We consider multi-access control for the uplink in OFDMA networks. Assuming that subcarriers are grouped into clusters, we investigate multichannel random access based on local channel state information, and propose an opportunistic multichannel Aloha scheme to maximize the system throughput. A key step is to build a mapping from a user's channel state information to its transmission probability and channel allocation. For the sake of comparison, we also characterize the throughput corresponding to the optimal centralized scheduling by using the Extreme-Value Theory of order statistics. We show that the opportunistic multichannel Aloha scheme is asymptotically order-optimal, in the sense that the only performance loss compared to the optimal centralized scheduling is due to the contention inherent in random access. In addition, we generalize the study to heterogeneous cases. Our findings show that when each user behaves as if it were in homogeneous systems, the proposed scheme can provide proportional fairness among the users.