Abstract Polylactide (PLA) has been widely used as a three-dimensional scaffold for cell transplantation and as a drug carrier for controlled release due to its superior biocompatibility and biodegradability. The technique of thermally induced phase separation (TIPS) is favorable for the fabrication of a porous scaffold due to its various advantages. A diverse range of porous structures can be obtained by TIPS through adjustment of process parameters. In this study, PLA scaffolds were prepared from a ternary PLA–dioxane–water system by the TIPS method. The effects of fabrication parameters, such as quenching route and the composition of the ternary system, on the macroscopic and microscopic morphologies of the scaffolds were studied. The phase diagram of this system showed that the presence of water non-solvent is essential for a liquid–liquid phase separation to occur. Accordingly, a binary PLA–dioxane system (100/0) exhibited only a solid–liquid phase separation. For a ternary PLA–dioxane–water system, the cloud-point temperature corresponding to a liquid–liquid phase separation was highly dependent on its water content and polymer concentration, while the freezing point corresponding to a solid–liquid phase separation was nearly independent of the polymer concentration and water content. Given a slight increase in the water content of the solvent mixture of 5%, i.e. a change in dioxane/water ratio from 90/10 to 85/15, or an increase in the polymer concentration of 2–30%, there was a substantial 40–50 °C increase in cloud-point temperature, in contrast with a nearly constant freezing point of between −10 and −20 °C. Morphology of the scaffold prepared via solid–liquid phase separation of a binary dioxane–PLA system gave a highly anisotropic tubular morphology with an internal ladder-like structure, while scaffolds prepared via liquid–liquid phase separation of a ternary PLA–dioxane–water system exhibited isotropic morphology. The ultimate pore structure of a scaffold was controlled by various processing parameters such as polymer concentration, water content, quench route and molecular weight of the polymer. Liquid–liquid de-mixing is a highly dynamic process that is subject to the effects of coarsening at the intermediate or later stage of the de-mixing process. For scaffolds prepared from solutions with polymer concentration of less than 7.5 wt%, the pore size decreases with the increase of polymer concentration. The coarsening effect was minimal and the final pore structure was found to be mainly determined by the liquid–liquid de-mixing mechanism, attributable to the smaller liquid–liquid de-mixing gap. For polymer concentrations higher than 7.5 wt%, the coarsening effect was more important, and a bimodal pore-size distribution was observed; in addition to the uniform small pores caused by instantaneous spinodal decomposition, large pores formed during the subsequent coarsening process. SEM micrographs showing that sedimentation occurred by aging time of about 5 and 60 min for scaffolds prepared from a 7.5 wt% PDLLA solution with dioxane/water ratio of 90/10 at −5 and −12 °C, respectively. Although this result suggests that such scaffolds rapidly lost macroscopic uniformity, porosity data indicates that the specimens did not lose integrity by aging for up to 120 min.