The paper presents phenix.refine, a program within the PHENIX package for crystallographic structure refinement. It supports refinement against experimental data with a wide range of resolution limits using a variety of model parameterizations. phenix.refine offers automation, flexibility, and extensive customization for complex cases. It allows multiple user-defined refinement strategies to be applied to specific model parts in a single run. An intuitive GUI is available for both novice and advanced users. X-ray or neutron diffraction data can be used separately or jointly. phenix.refine is tightly integrated into the PHENIX suite, serving as a critical component in automated model building, final structure refinement, structure validation, and deposition to the wwPDB. Crystallographic structure refinement involves a complex process with diverse steps, each requiring decisions on model parameterization, refinement target, and optimization method. These decisions are influenced by data quality and model quality. Model parameters include atomic model parameters (atomic coordinates, ADPs, occupancies, anomalous scattering terms) and non-atomic parameters (bulk solvent, twinning, crystal anisotropy). The total model structure factors $ F_{model} $ are expected to match observed values $ F_{obs} $ and other experimental data. Refinement targets are mathematical functions that quantify the fit of model parameters to experimental data. The goal of refinement is to minimize the target function, which is achieved through various optimization methods, including gradient-driven minimization, simulated annealing, and interactive model building. phenix.refine supports a wide range of refinement strategies for different data resolutions, from ultrahigh to low resolution. It includes tools for neutron data refinement, such as automatic detection and building of exchangeable H/D sites and difference electron-density map-based building of D atoms for water molecules. Refinement can be performed in real or reciprocal space or consecutively in both. Refinement against twinned data is also possible. The flexibility and functionality of phenix.refine are enabled by modern software development approaches, including object-oriented languages and library-based development. The refinement protocol in phenix.refine consists of three main parts: initialization, macro-cycle, and output. The initialization includes processing input data, job-control parameters, and strategy selection. The macro-cycle is the main body of refinement, where the actual model refinement occurs. The output step reports the refined model, electron-density maps, and statistics. Key steps in structure refinement include processing inputs, total model structure factor refinement, ordered solvent modeling, refinement targets and weights, coordinate refinement, ADP refinement, occupancy refinement, and refinement of dispersive and anomalous coefficients. The refinement output includes PDB, LOG, MTZ, GEO, and EFF files, as well as various map calculations. phenix.refine also includes specific tools for refinement at subatomic and low resolutions. At subatomic resolution, tools are available for model completion and refinement, including unrestrained coordinate and ADP refinement, IAS models, H atomThe paper presents phenix.refine, a program within the PHENIX package for crystallographic structure refinement. It supports refinement against experimental data with a wide range of resolution limits using a variety of model parameterizations. phenix.refine offers automation, flexibility, and extensive customization for complex cases. It allows multiple user-defined refinement strategies to be applied to specific model parts in a single run. An intuitive GUI is available for both novice and advanced users. X-ray or neutron diffraction data can be used separately or jointly. phenix.refine is tightly integrated into the PHENIX suite, serving as a critical component in automated model building, final structure refinement, structure validation, and deposition to the wwPDB. Crystallographic structure refinement involves a complex process with diverse steps, each requiring decisions on model parameterization, refinement target, and optimization method. These decisions are influenced by data quality and model quality. Model parameters include atomic model parameters (atomic coordinates, ADPs, occupancies, anomalous scattering terms) and non-atomic parameters (bulk solvent, twinning, crystal anisotropy). The total model structure factors $ F_{model} $ are expected to match observed values $ F_{obs} $ and other experimental data. Refinement targets are mathematical functions that quantify the fit of model parameters to experimental data. The goal of refinement is to minimize the target function, which is achieved through various optimization methods, including gradient-driven minimization, simulated annealing, and interactive model building. phenix.refine supports a wide range of refinement strategies for different data resolutions, from ultrahigh to low resolution. It includes tools for neutron data refinement, such as automatic detection and building of exchangeable H/D sites and difference electron-density map-based building of D atoms for water molecules. Refinement can be performed in real or reciprocal space or consecutively in both. Refinement against twinned data is also possible. The flexibility and functionality of phenix.refine are enabled by modern software development approaches, including object-oriented languages and library-based development. The refinement protocol in phenix.refine consists of three main parts: initialization, macro-cycle, and output. The initialization includes processing input data, job-control parameters, and strategy selection. The macro-cycle is the main body of refinement, where the actual model refinement occurs. The output step reports the refined model, electron-density maps, and statistics. Key steps in structure refinement include processing inputs, total model structure factor refinement, ordered solvent modeling, refinement targets and weights, coordinate refinement, ADP refinement, occupancy refinement, and refinement of dispersive and anomalous coefficients. The refinement output includes PDB, LOG, MTZ, GEO, and EFF files, as well as various map calculations. phenix.refine also includes specific tools for refinement at subatomic and low resolutions. At subatomic resolution, tools are available for model completion and refinement, including unrestrained coordinate and ADP refinement, IAS models, H atom