We determined the relationship of forest composition and alpha diversity (species diversity of a local assemblage) to altitude, soil and spatial factors over a 440 - 2950 ma.s.l gradient. Altitudinal gradient on the Atlantic slope of the Talamanca Cordillera, Costa Rica. We measured and identified all stems > 10 cm dbh in 32 0.25-ha undisturbed rain forest plots over the gradient. We determined compositional patterns using Non-Metric Multidimensional Scaling (NMS) ordination, and used linear regressions to explore the relationship between four alpha diversity metrics and altitude. With variation partitioning (VARPART) we determined the compositional variation explained by altitude, soil and spatial variables quantified using Principle Components of Neighbour matrices. We identified 425 species. NMS axis 1 separated a lowland zone (440-1120 m asl) from a transitional one dominated by holarctic Oreomunnea mexicana (1400-1600 m asl) and Quercus-dominated forests at altitudes >2100 m asl. The lowland zone was separated into two clusters of plots on NMS axis 2, the first in the 430-620 m asl range and the second at 1000-1120 masl. . Regressions showed that alpha diversity metrics were strongly negatively related to altitude (R2>0.78). Overall adjusted R2 from VARPART was 0.43, with 0.30, 0.21 and 0.17 for altitude, soil and space respectively. The respective adjusted R2 of individual matrices, on controlling for the other two, were 0.06, 0.05 and 0.09 (p<0.001). There are two well-defined forest compositional zones on this gradient – lowlands and montane forests – with a transitional zone at 1400-1600 m asl where lowland tropical and montane holarctic species are found together. Montane forests are very distinct in their composition and low alpha diversity. Vegetation and soil respond to altitude, and therefore temperature, as an integrated system, a model that goes beyond niche assembly as shown by the significant effect of space in the VARPART.