The scale-similarity model in large-eddy simulation (LES) leads to an attractive, functionally simple expression for the subgrid-scale (SGS) stress tensor. It is well known, however, that the similarity model fails to accurately predict some of the most fundamental quantities in turbulent flows, perhaps the most important being the global energy transfer and the associated subgrid-scale dissipation. To address this, additional dissipative terms are usually added to the similarity model to improve its performance. In the present paper, considerations of interscale energy transfer have been used to identify sources of the observed deficiencies of the similarity model, specifically its inadequate balancing of terms contributing energy to the smallest scales and its duplication of terms producing effects in the largest scales. These considerations provide guidance in the development of a new model, which shows more favorable characteristics of energy transfer while preserving the functional simplicity of the scale-similarity model. Partial nonlinear terms are used to decompose the nonlinear transfer present in LES and to formulate a model expression capable of balancing small-scale production terms depositing energy near the LES cutoff. The proposed model is formulated in the same vein as the scale-similarity model, consisting of test filtered velocities and their products, but offers clear improvements in predictions of mean flow quantities and the global energy flux from the resolved to subgrid scales without the need for additional terms to augment subgrid-scale energy dissipation. The application of the new interscale transfer model in LES of wall-bounded flows leads to predictions of mean and RMS flow quantities comparable to those obtained for other, established SGS models.