There has been a growing interest in studying and modelling unsaturated soil behaviour over the last 15 years. It is generally believed that soil suction affects dilatancy of unsaturated soils and some suction-dependant dilatancy relationships are assumed in many flow rules of existing elasto-plastic models. However, limited laboratory data reported in literature explicitly illustrate how suction affects dilatancy of an unsaturated soil during shear. In this paper, a series of laboratory tests on a compacted completely decomposed granite (CDG) were carried out in a suction-controlled direct shear box. Five suctions varying from 0 to 400 kPa were considered. Each specimen was wetted to close to its full saturation and then dried to a pre-determined soil suction before shear. During shear, each specimen was subjected to a constant net vertical stress and suction. Effects of soil suction on the evolution of dilatancy, stress dilatancy relationship and maximum dilatancy are presented and discussed in this paper.