To determine whether an intervention that improves nursing home (NH) staff connections, communication, and problem solving (CONNECT) would improve implementation of a falls reduction education program (FALLS). Cluster randomized trial. Community (n=4) and Veterans Affairs (VA) NHs (n=4). Staff in any role with resident contact (n=497). NHs received FALLS alone (control) or CONNECT followed by FALLS (intervention), each delivered over 3 months. CONNECT used storytelling, relationship mapping, mentoring, self-monitoring, and feedback to help staff identify communication gaps and practice interaction strategies. FALLS included group training, modules, teleconferences, academic detailing, and audit and feedback. NH staff completed surveys about interactions at baseline, 3 months (immediately after CONNECT or control period), and 6 months (immediately after FALLS). A random sample of resident charts was abstracted for fall risk reduction documentation (n=651). Change in facility fall rates was an exploratory outcome. Focus groups were conducted to explore changes in organizational learning. Significant improvements in staff perceptions of communication quality, participation in decision-making, safety climate, caregiving quality, and use of local interaction strategies were observed in intervention community NHs (treatment-by-time effect P=.01) but not in VA NHs, where a ceiling effect was observed. Fall risk reduction documentation did not change significantly, and the direction of change in individual facilities did not relate to observed direction of change in fall rates. Fall rates did not change in control facilities (falls/bed per year: baseline, 2.61; after intervention, 2.64) but decreased by 12% in intervention facilities (falls/bed per year: baseline, 2.34; after intervention, 2.06); the effect of treatment on rate of change was 0.81 (95% confidence interval=0.55-1.20). CONNECT has the potential to improve care delivery in NHs, but the trend toward improving fall rates requires confirmation in a larger ongoing study.