This paper proposes a new approach to the optimal allocation of airline seats among itineraries in a network under stochastic demand. Current approaches to network seat inventory control are based on either full assignment or bucket control. Under full assignment, all seats on each flight leg are assigned to a specific single- or multi-leg itinerary. Under bucket control seats are assigned to groups of itineraries. The flexible assignment approach proposed in this paper assigns some seats exclusively to each single- or multi-leg itinerary as in fixed assignment and assigns the remaining seats to groups of itineraries as in bucket control. This paper formulates the flexible assignment approach, develops rules for optimal assignment for a single-fare-class network and proves the superiority of this approach to either fixed assignment or bucket control. A numerical simulation is ised to illustrate the differences in expected revenue among these three methods of seat inventory control for four different demand scenarios.