The objective of this work was to identify, through the use of multivariate analysis, the spontaneous accesses of castor bean that show promising characteristics for inclusion in genetic improvement programs of this culture in Brazil. The study was conducted at the Paulista State University, Jaboticabal Campus. The accesses of castor bean seeds obtained were identified as ACS-001-CRSP, ACS-001-JASP, ACS-001-MASP, ACS-001-SESP, ACS-002-JASP, ACS-002-SESP and ACS-003-JASP, from the municipalities of Jaboticabal, Monte Alto, Santa Ernestina and from the District of Córrego Rico. The seeds' biometric variables and germination potential were measured, in addition to the biomass accumulation of the seedling's vegetative parts. The experimental design was completely randomized, with seven treatments and four replications. The original data was standardized and submitted to multivariate analysis of principal components and grouping. Through the results obtained, it is possible to reduce the number of original variables into two latent variables with discriminant power between the castor bean accesses, being the multivariate analysis efficient in this process. The castor bean accesses ACS-001-CRSP and ACS-001-MASP are promising for introduction in genetic improvement programs of this culture.