The excitation to the LPC filter in CELP based coders includes two components: the contribution of the stochastic codebook and the contribution of the pitch or long-term predictor, which can be derived from an adaptive codebook in which the codewords are delayed versions of the past excitation signal. This paper is devoted to a systematic study of codebook related issues in CELP based coders: open and closed-loop schemes for determining the optimum adaptive and stochastic codewords, dimension and overlap between analysis frames, codebook population, etc. We shall cover some of the most well known search procedures and introduce two new variants that significantly reduce the computational complexity of the stochastic search procedure, without quality degradation: the truncated autocorrelation approach and the unity-magnitude approach.