We give definitions on increasing, constant and decreasing returns to scale under thegeneralized data envelopment analysis (GDEA) model. The GDEA model possesses cone structures fordescribing evaluators' preferences toward different decision making units (DMU) and toward differentinput/output categories. The GDEA model contains many important DEA models as its subclasses.Based upon the definition, we discuss and rigorously prove the necessary and sufficient conditionsfor a DMU to have increasing, constant and decreasing returns to scale under a GDEA model, anddiscuss and clarify the differences and relationships of these conditions under different subclasses ofthe GDEA model. Furthermore, we describe a method based on a two-stage linear program to testreturns to scale properties.