Particle swarm optimization (PSO) is a population-based stochastic search algorithm that has been widely used to solve many real-world problems. However, like other evolutionary algorithms, PSO also suffers from premature convergence and entrapment into local optima when addressing complex multimodal problems. In this paper, we propose a chaos-embedded particle swarm optimization algorithm (CEPSO). In CEPSO, the chaos-based swarm initialization is first applied to yield high-quality initial particles with better stability. Afterwards the chaotic inertia weight and the chaotic sequence based random numbers are introduced into the velocity update scheme for PSO to improve its global and local search capabilities. In addition, two different mutation strategies (chaos and levy) are utilized to enhance the swarm diversity without being trapped in local optima. Finally, the CEPSO proposed in this work is compared with several classical PSOs on a set of well-known benchmark functions. Experimental results show that CEPSO can achieve better performance compared to several other PSO variants in terms of the solution accuracy and convergence rate.