The paper deals with the system dynamics modeling of a stochastic behavior. The starting point is replacing the traditional system dynamics model with a discrete-time stochastic dynamic model in which state variables are measured indirectly, through noisy and incomplete measurements. The state variables and possible unknown parameters in such a model can be systematically estimated from the available measurements using the Bayesian paradigm. Closed-form solutions exist only for few special cases, such as a linear normal model with known parameters, otherwise numerical approximations are required. The paper suggests a particle filter algorithm as a particularly appealing approximation that preserves much of the intuitive workings of system dynamics. A practical example illustrates both the stochastic modeling process and the approximate Bayesian analysis.