Microencapsulation was studied by a thermodynamic approach taking quantitatively into account the molecular interactions between polymer, solvent and the aqueous protein phase. Bovine serum albumin was microencapsulated into poly(D,L-lactic acid) by spray drying using single solvents and binary solvent mixtures. The use of binary solvent mixtures allowed systems with adjustable solubility parameters to be designed. Microsphere characteristics like entrapment efficiency, burst release after 24 h and surface morphology were investigated and proved to be highly dependent on the polymer solvent system. Hildebrand or partial Hansen solubility parameters (delta, delta(d), delta(p), delta(h)) proved to be insufficient for predicting microsphere properties, although low or moderate hydrogen-bonding solvents and solvent mixtures were found to be generally appropriate, whereas strongly I-I-bonding solvents gave poor quality microspheres; Similarly, water miscible solvents were shown to give often unsatisfactory products. A more powerful tool for optimizing a microencapsulation process is estimating polymer-solvent-drug interactions by using delta(d), delta(p), and the Drago parameters E (electrostatic) and C (covalent) of the components involved. Entrapment efficiency is increased and burst release reduced if polymer-drug interaction is dominant and polymer-solvent, drug-solvent interactions are reduced. This thermodynamic approach provides a rational basis for optimizing microencapsulation processes.