Texture analysis is an important approach in textile quality control. Higher order statistics have been very useful in problems where non-Gaussianity, nonminimum phase, colored noise or nonlinearity is important. In this work, higher order statistical analysis is applied to texture defect detection problem. A neighborhood definition is proposed for cumulant lags of higher order statistics and it is used to form higher order statistical feature sets. These higher order statistical feature sets and some hybrid feature sets composed of both second and higher order statistics are used to detect defects on textural images of textile fabrics. The results are compared with methods based only on second order statistics both from performance and computational complexity points of view.