This article aims to link the mainstream subject of chain-folded polymer crystallization with the rather speciality field of extended-chain crystallization, the latter typified by the crystallization of polyethylene (PE) under pressure. Issues of wider generality are also raised for crystal growth, and beyond for phase transformations. The underlying new experimental material comprises the prominent role of metastable phases, specifically the mobile hexagonal phase in polyethylene which can arise in preference to the orthorhombic phase in the phase regime where the latter is the stable regime, and the recognition of "thickening growth" as a primary growth process, as opposed to the traditionally considered secondary process of thickening. The scheme relies on considerations of crystal size as a thermodynamic variable, namely on melting-point depression, which is, in general, different for different polymorphs. It is shown that under specifiable conditions phase stabilities can invert with size; that is a phase which is metastable for infinite size can become the stable phase when the crystal is sufficiently small. As applied to crystal growth, it follows that a crystal can appear and grow in a phase that is different from that in its state of ultimate stability, maintaining this in a metastable form when it may or may not transform into the ultimate stable state in the course of growth according to circumstances. For polymers this intermediate initial state is one with high-chain mobility capable of "thickening growth" which in turn ceases (or slows down) upon transformation, when and if such occurs, thus "locking in" a finite lamellar thickness. The complete situation can be represented by a P, T, 1/l (l ≡ crystal thickness) phase-stability diagram which, coupled with kinetic considerations, embodies all recognized modes of crystallization including chain-folded and extended-chain type ones. The task that remains is to assess which applies under given conditions of P and T. A numerical assessment of the most widely explored case of crystallization of PE under atmospheric pressure indicates that there is a strong likelihood (critically dependent on the choice of input parameters) that crystallization may proceed via a metastable, mobile, hexagonal phase, which is transiently stable at the smallest size where the crystal first appears, with potentially profound consequences for the current picture of such crystallization. Crystallization of PE from solution, however, would, by such computations, proceed directly into the final stage of stability, upholding the validity of the existing treatments of chain-folded crystallization. The above treatment, in its wider applicability, provides a previously unsuspected thermodynamic foundation of Ostwald's rule of stages by stating that phase transformation will always start with the phase (polymorph) which is stable down to the smallest size, irrespective of whether this is stable or metastable when fully grown. In the case where the phase transformation is nucleation controlled, a ready connection between the kinetic and thermodynamic considerations presents itself, including previously invoked kinetic explanations of the stage rule. To justify the statement that the crystal size can control the transformation between two polymorphs, a recent result on 1 -4-poly-trans-butadiene is invoked. Furthermore, phase-stability conditions for wedge-shaped geometries are considered, as raised by current experimental material on PE. It is found that inversion of phase stabilities (as compared to the conditions pertaining for parallel-sided systems) can arise, with consequences for our scheme of polymer crystallization and with wider implications for phase transformations in tapering spaces in general. In addition, in two of the Appendices two themes of overall generality (arising from present considerations for polymers) are developed analytically; namely, the competition of nucleation-controlled phase growth of polymorphs as a function of input parameters, and the effect of phase size on the triple point in phase diagrams. The latter case leads, inter alia to the recognition of previously unsuspected singularities, with consequences which are yet to be assessed. © 1994 Chapman & Hall.