Abstract The vegetation of shell-beds is studied in an island on the coast of Skagerak, SE Norway. Within this restricted area 125 sample plots, 1 m2 each, were distributed by a restricted random procedure. Frequency in subplots and percentage cover were used for estimation of species abundance. A set of 33 environmental variables was obtained from each sample plot. Ecological gradients corresponding to the main vegetational gradients are identified by means of DCA ordination results correlated with environmental data. DCA ordination results are further interpreted by the use of an earlier classification of the vegetation and by application of Grime’s strategy concept. The following complex gradients were recognized: ( 1) the primary successional gradient, influenced by variables dependent on site age, e.g. elevation, distance from the sea, and rate of decomposition of mollusc shells, (2) the secondary successional gradient, from open meadow vegetation via open shrub vegetation to dense shrub vegetation, highly correlated with variables independent of site age, e.g intensity of grazing, (3) the nutrient availability gradient, and (4) the soil moisture gradient. Interpretation of relations between coenoclines and environmental variables was performed by Kendall rank correlation and relationships among ecological variables analysed by PCA ordination. Direct gradient analysis by rh-DCCA is used as a supplement to DCA. The ratio between the rh-DCCA eigenvalue of the first ordination axis and the corresponding eigenvalue of the first residual (unconstrained) axis is used to quantify the importance of each environmental variable. The performed multivariate analyses show that the successional pathways on shell-bed shores are dependent on the grazing pressure. The relative importance of the following structuring processes is discussed: (1) interspecific interaction (competition, coexistence and facilitation), (2) destabilizing factors (disturbance and fluctuations), (3) stress, and (4) chance. Ordination and association analyses shows that Juniperus communis is the most important structuring species in the material. Grazing is necessary to reduce overgrowth by junipers and to maintain a high species diversity. The field data are used for inspection of the performance of different rescaling and detrending options, available with DCA ordination. Non-linear rescaling appears to be less influenced by deviant sample plots as compared to linear rescaling, and the S.D. units of the axes produced by non-linear rescaling are considered better interpretable in terms of a unimodal model of species responses to ecological gradients. Available options for detrending by polynomials are shown to be inferior to the detrending by segments option, as polynomial detrending may impose a polynomial distortion on the point configuration in the ordination space.