While the emergence and spread of antibiotic resistance has been well studied for endemic infections, comparably little is understood about this problem in the context of epidemic infections such as influenza. The availability of antimicrobial treatments for epidemic diseases raises the urgent question of how to deploy treatments in a population to achieve maximum benefit despite resistance evolution. Recent simulation studies have shown that the number of cases prevented by antimicrobials can be maximized by optimally delaying the use of treatments during an epidemic.  These studies focus on the indirect benefit of antimicrobial use: preventing disease among untreated individuals. Here we identify and examine the direct benefit of antimicrobial use, namely, the number of successfully treated cases.  We develop mathematical models to study how  the schedule of antiviral use influences the success or failure of subsequent use due to the spread of resistant strains. The direct benefit of drugs is maximized when their use is postponed, even if an unlimited supply of the drug is available.  This occurs because the early use of antimicrobials disproportionately drives emergence and spread of antibiotic resistance, leading to subsequent treatment failure. However, in the case of antimicrobials with low effect on transmission, the benefits of delaying antimicrobial deployment are only modest and can only be reaped if the trajectory of the epidemic can be accurately estimated early. Given the uncertainties faced in most epidemic situations it will usually be sensible to initiate widespread antimicrobial use as early as possible.