The Bees Algorithm (BA) is a recent population-based optimization algorithm, which tries to imitate the natural behavior of honey bees in food foraging. This meta-heuristic is widely used in various engineering fields. However, it suffers from certain limitations. This paper focuses on improvements to the BA in order to improve its overall performance. The proposed enhancements were applied alone or in pair to develop enhanced versions of the BA. Three improved variants of BA were presented: BAMS-AN, HBAFA and HFBA. The new BAMS-AN includes memory scheme in order to avoid revisiting previously visited sites and an adaptive neighborhood search procedure to escape from local optima during the local search process. HBAFA introduces the Firefly Algorithm (FA) in local search of BA to update the positions of recruited bees, thus increasing exploitation in each selected site. The third improved BA, ie HFBA, employs FA to initialize the population of bees in the BA for a best exploration and to start the search from more promising regions of the search space. The proposed enhancements to the BA have been tested using several continuous benchmark functions and the results have been compared to those achieved by the standard BA and other optimization techniques. The experimental results indicate that the improved variants of BA outperform the standard BA and other algorithms on most of the benchmark functions. The enhanced BAMS-AN performs particularly better than others improved BAs in terms of solution quality and convergence speed.