Three algorithms for global energy minimization based on the simulated annealing of the classical density distribution are presented. The first is based on annealing the classical density distribution directly in temperature and is the classical analog of imaginary time quantum dynamics. Another two algorithms are based on the approximate solution of the classical Liouville equation for the dynamics of a system coupled to a heat bath using a rigid temperature constraint and Fokker-Planck dynamics. These three methods are compared with standard simulated annealing based on molecular dynamics. The results for a series of Lennard-Jones clusters demonstrate that by annealing the continuous density distribution (representing a volume of phase points) the Likelihood of finding the global minimum is dramatically enhanced.