Journal of Applied Sciences1812-56541812-5662Asian Network for Scientific Information10.3923/jas.2007.1626.1632Hamza El-SayedMohamedAhmed M. Khedr122007712A hidden Markov model for protein modelling consists of sub-models for alpha-helix, beta-sheets, coil and possibly more. It is described how to estimate the model parameters as a whole from labeled sequences instead of estimating the parameters from the individual parts independently from subsequences. It is argued that the standard maximum likelihood ML estimation is not the optimal for training such model. In this study a new method is used where instead of estimating the parameters of model that maximizing the probability of the protein sequences (ML), we maximize the probability of the correct labels prediction, such a criterion is called conditional maximum likelihood CML. The advantage of this method is to optimize recognition of model. We tested our method on some of protein families such as L-asparagines, we noted that the performance of HMM is improved in prediction process.]]>Anders, K., M. Brown and I.S. Mian,1994Anders, K.,1994Kotzia, G.A. and N.E. Labrou,2005Masuo, N. and K. Yoshimune,2004Micrococcus luteus K-3-type glutaminase from Aspergillus oryzae RIB40.]]>Normandin, Y. and S. Morgan,1991Rabinar, J.,1991Rabiner, L.,1989Riis, S.K. and A. Krogh,1997Yoshimune, K., N. Esaki and M. Moriguchi,2005