Pontraygin's maximum principle (MP) involving the Hamiltonian system and Bellman's dynamic programming (DP) involving the HJB equation are the two most important approaches in modern optimal control theory. However, these two approaches have been developed separately in literature and it has been a long-standing, yet fundamentally important problem to disclose the relationship between them and to establish a unified theory. The problem is by no means a "new" one; indeed, it roots in the Hamilton-Jacobi theory in analytic mechanics and method of characteristics in classical PDE theory, and has intrinsic relation with the Feynman-Kac formula in stochastic analysis and shadow price theory in economics. This paper discusses some deep connections between the MP and DP in stochastic optimal controls from various aspects.