This study evaluates the relative robustness of various ordination techniques using simulated vegetation data. The methods compared include local non-metric multidimensional scaling (LNMDs), detrended correspondence analysis (DCA), Gaussian ordination (GO), principal components analysis (PCA), and principal co-ordinates analysis (PCoA). The results indicate that linear techniques (PCA, PCoA) are ineffective due to curvilinear distortion, while GO is sensitive to noise and not robust to departures from a symmetric, unimodal response model. DCA also lacks robustness to variations in the response model and sampling pattern, often showing significant distortions in two-dimensional models. LNMDs are recommended as a robust technique for indirect gradient analysis, suggesting they deserve more widespread use by community ecologists. The study uses an early version of COENOS to generate artificial data matrices, examining model properties such as the number of underlying gradients, the shape of species' ecological responses, and the arrangement of sites in the simulated environment space.This study evaluates the relative robustness of various ordination techniques using simulated vegetation data. The methods compared include local non-metric multidimensional scaling (LNMDs), detrended correspondence analysis (DCA), Gaussian ordination (GO), principal components analysis (PCA), and principal co-ordinates analysis (PCoA). The results indicate that linear techniques (PCA, PCoA) are ineffective due to curvilinear distortion, while GO is sensitive to noise and not robust to departures from a symmetric, unimodal response model. DCA also lacks robustness to variations in the response model and sampling pattern, often showing significant distortions in two-dimensional models. LNMDs are recommended as a robust technique for indirect gradient analysis, suggesting they deserve more widespread use by community ecologists. The study uses an early version of COENOS to generate artificial data matrices, examining model properties such as the number of underlying gradients, the shape of species' ecological responses, and the arrangement of sites in the simulated environment space.