The Karhunen-Loeve transform (KLT) is an optimal method for dimensionality reduction, widely applied in image compression, reconstruction and retrieval, pattern recognition and classification. The basic idea consists in evaluating, starting from a set of representative examples, a reduced space, which takes into account the structure of the data distribution as much as possible, and representing each element in such an uncorrelated space. Unfortunately, KLT has the drawback of requiring a periodical recomputation in presence of a dynamic dataset. This work presents a novel efficient approach to merge multiple eigenspaces, which provides an incremental method to compute an eigenspace model by successively adding new sets of elements. Experimental results show that the merged model grants performances as good as a one obtained by a batch procedure.