Recent studies have provided evidence that distance-deterrence parameter s may vary between areas even in doubly constrained spatial interaction models. In particular, it has been suggested that distance effects may be stronger for less accessible origins or destinations. This paper considers possible explanations for such variation. It is argued that explanations in terms of `map pattern' alone are inadequate and that more substantive explanations are required in relation to the economic assumptions of the standard spatial interaction model. Four potentially variable parameters in this model are distinguished and three specific hypotheses to account for the observed pattern are outlined. These involve a real income effect, scale economies in transport costs, and the spatial concentration of specialised functions.