In this study, we have applied percentage discounting technique to overcome a drawback in Good- Turing smoothing technique. Data sparseness is an inherent and a severe problem in language modeling. Smoothing is one of the important processes to handle this problem. There are several well-known smoothing techniques which are used to solve data sparseness problem. In general, smoothing techniques ignore linguistic knowledge and are particularly based on statistical hypotheses. Good Turing is very effective for data sparseness problem but it has a drawback that it calculates zero probability if frequency of next frequency is zero. Consequently a new technique is presented in this study, which is percentage discounting technique and this technique can overcome the drawback of Good Turing smoothing.