BACKGROUND Modifiable unit characteristics, including nurse work environment, education, certification, and staffing have been shown to impact patient safety. Physical restraints are an important patient safety issue, however the relationships between these modifiable unit characteristics and physical restraint use on critical care units has not been explored. OBJECTIVE Our objective was to determine the role of nursing work environment, nurse education and certification, and nurse staffing on physical restraint use. METHODS The study was a secondary analysis of 2017-2018 unit-level restraint rates from the National Database of Nursing Quality Indicators linked to Registered Nurse survey and hospital characteristics data. METHODS Work environment was examined at the unit-level using total Practice Environment Scale of the Nursing Work Index and subscale scores. Unit-level nurse expertise included the percent of nurses with at least a Bachelor of Science in Nursing degree and percent with certification. Nurse staffing was the unit-level Registered Nurse hours per patient day and skill mix. Analyses included descriptive statistics, bivariate correlations, and Structural Equation Modeling. We used a first order model to estimate predicted restraint use from the five subscales of the Practice Environment Scale of the Nursing Work Index. A higher order model predicted restraint use from a total work environment factor score. RESULTS The sample included 408 critical care units from 226 hospitals. Mean restraint rate was 15% (SD 12, range 0-53%). In the first-order model, the Collegial Nurse-Physician Relations subscale had a positive relationship with restraint use (β = 0.167, 95% confidence interval 0.010-0.333). In the higher order model, the total work environment score had a negative relationship with restraint use (β = -0.088, 95% confidence -0.178- -0.014). Registered Nurse hours per patient day had a negative relationship to restraint use in both the first order (β = -0.114, 95% confidence interval -0.222--0.025) and higher order models (β = -0.117, 95% confidence interval -0.223- -0.012). CONCLUSIONS We found that better nurse-physician relationships were associated with higher restraint use. This finding is supported by previous literature and may reflect physician trust in nursing judgement when using restraints. However, a better work environment overall was associated with lower restraint rates. Further, Registered Nurse hours per patient day, but not skill mix, was associated with lower restraint rates. We conclude that improving the overall nurse work environment and nurse staffing, as well as using interprofessional interventions, may be successful in decreasing restraint use on critical care units.