Multivariate cluster and canonical variate analyses were undertaken for 10 genotypes of linseed (Linum usitatissimum L.) that were tested in a four-times replicated randomized block design across 18 environments (six localities by 3 years) of Ethiopia. The main aims of this study were to determine the similarities and differences of the genotypes and their testing environments, and to compare applicability of the two statistical methods. Cluster analysis grouped the genotypes into five classes in accordance with their original sources. The six locations and 18 environments were stratified into four and seven clusters, respectively. Three sites (Bekoji, Kulumsa and Sinana) were separately stratified, while three other ones (Holetta, Asasa and Adet) showed closer similarity. Canonical variate analysis indicated that ‘D33C’ and ‘D24C’ were distinguished from the other genotypes by their high oil contents. ‘N10D’ and ‘Norlin’ had closer values and were thus preferred for their good seed yield and earliness. Days to flowering and maturity, oil contents and lodging per cent played major roles in discriminating the genotypes. Comparison of the two methods showed clearer differentiation by cluster analysis than canonical variate analysis. Canonical variate analysis also contributed information on how each variable discriminated the genotypes and their test environments. Thus, both methods complement each other in providing useful information for more efficient variety development programmes.