Bayes net is DAG (nodes as rvs) together with the conditional likelihoods at each node: n/pa(n). Usual parameterization is induced by modeling each conditional using (conditional) exponential model.

Markov Net is undirectedgraph (nodes as rvs) together with set of factors. graph+no.factors+scopeOfEach is called network structure. No restrictions on number and factor functions; but scope has important link with graph: every scope must form clique. Unlike in Bayes nets there is no bijection between DAG and network structure. Usual parameterization is exp(\sum_i w_i phi_i(S_i)/Z(w), where phi_i is i^{th} clique potential (and exp(w_i\phi_i(S_i) is i^{th} factor).