Nonribosomal peptide synthetases (NRPSs) catalyze the assembly of a large number of complex peptide natural products, many of which display therapeutically useful activity. Each cycle of chain extension is carried out by a dedicated module of the multifunctional enzymes. A module harbors all the catalytic units, which are referred to as domains, necessary for recognition, activation, covalent binding, and optionally modification of a single building block monomer, as well as for peptide-bond formation with the growing chain. A terminal domain releases the full-length peptide chain from the enzyme complex. Recent characterization of many NRPS systems revealed several examples where the sequence of the product does not directly correspond to the linear arrangement of modules and domains within the enzyme(s). It is now obvious that these systems cannot be regarded as rare exceptions of the common NRPS architecture but rather represent more complicated variations of the NRPS repertoire to increase their biosynthetic potential. In most of these cases unusual peptide structures of the products are observed, such as structures with side-chain acylation, cyclization involving the peptide backbone and/or side chains, and transfer of the peptide chain onto soluble small-molecule substrates. These findings indicate a previously unexpected higher versatility of the modules and domains in terms of both catalytic potential and interaction within the multifunctional protein templates. We propose to classify the known NRPS systems into three groups, linear NRPSs (type A), iterative NRPSs (type B), and nonlinear NRPSs (type C), according to their biosynthetic logic. Understanding the various biosynthetic strategies of NRPSs will be crucial to fully explore their potential for engineered combinatorial biosynthesis.