K-mean algorithm requires total number cluster, k beforehand in order the algorithm operates correctly. This pre-requisite value is needed to ensure the algorithm works on the tested data. In this paper, a test-and-generate approach is applied to estimate total number present in a data. A hybrid Bees Algorithm and cluster validity index are used for this purpose. The modified Bees algorithm is used to find near-optimal cluster centres (centroids) whereas cluster validity index is employed to examine `goodness' of the generated clusters. A series of experiments using some benchmarking data sets are undertaken to evaluate effectiveness of the proposed approach. A promising results show that the proposed approach is capable to estimate total number of cluster in a data.