We consider a cognitive radio scenario where the communication between two secondary users (SUs) exploits opportunistic spectrum access (OSA) over a wireless channel licensed to primary users (PUs). Assuming a slotted MAC over a single frequency channel and imperfect spectrum sensing, we address the problem of determining the packet size and Forward Error Correction (FEC) coding rate that maximize the SU communication goodput, i.e., the amount of payload bits correctly received in the unit time. Assuming a Markovian model for the PU activity, a saturation regime for the SU, and periodic channel sensing, we find out the mean time between two consecutive packet transmissions and derive approximate analytical expressions, in closed form, which provide upper and lower bounds on the SUs goodput. Such expressions show the dependence of the SU goodput on packet size, FEC coding rate, signal to noise plus interference ratio, amount of resources allocated to sensing, packet overhead, and primary traffic statistics. We provide simulation results showing that the derived analytical expressions are very close to the actual system performance. We also evaluate the sensitivity of the optimal packet size and FEC coding rate pair to the operational conditions represented by the received power and the PU traffic load, showing that the optimal pair is very sensitive to the former, and only moderately affected by the latter.