The traditional parameter used as a measure of vulnerability of a network modeled by a graph with perfect nodes and edges that may fail is edge connectivity lambda. For the complete bipartite graph K-p,K-q where 1 <= p <= q, lambda(K-p,K-q) = p. In this case, failure of the network means that the surviving subgraph becomes disconnected upon the failure of individual edges. If, instead, failure of the network is defined to mean that the surviving subgraph has no component of order greater than or equal to some preassigned number k, then the associated vulnerability parameter, the k-component order edge connectivity lambda((k))(c), is the minimum number of edges required to fail so that the surviving subgraph is in a failure state. We determine the value of lambda((k))(c)(K-p,K-q) for arbitrary 1 <= p <= q and 4 <= k <= p + q. As it happens, the situation is relatively simple whenp is small and more involved whenp is large.