The design of current acoustic-phonetic decoders for a specific language involves the selection of an adequate set of sub-lexical units. In previous works [1] a set of context independent units was presented for Spanish Continuous Speech Recognition. The aim of this work was to extend this basic set representing context variability under a discriminative-transitional/steady criterion. Our main goal was to obtain a good trade-off between discriminative ability, context variability and number of units. Finally, and within the framework of Hidden Markov Modelling, different series of acoustic-phonetic decoding experiments were earned out over a Spanish Continuous Speech corpus. The phone recognition rates obtained through these experiments showed that the proposed criterion seems to be an interesting approach to model context v ariability in Spanish.