This paper addresses the issue of extracting textural feature for timber defect detection. Statistical features based on spatial dependence matrix are extracted for both classes; clear wood and defect. Instead of using the classical directional matrices, rotation invariant spatial dependence matrix formulation is applied to ensure accurate detection regardless of the timber feed direction. Hotelling T-Squared test is used to measure significance difference of mean between feature distributions of the two classes. The result will give some indication to whether the features extracted are sufficient/good enough to be used in future classification stage.