A bibliometric analysis based on the Science Citation Index-Expanded (SCI-Expanded) database from the Web of Science was performed to review urban heat island (UHI) research from 1991 to 2015 and statistically assess its developments, trends, and directions. In total, 1822 papers published in 352 journals over the past 25 years were analyzed for scientific output; citations; subject categories; major journals; outstanding keywords; and leading countries, institutions, authors, and research collaborations. The number of UHI-related publications has continuously increased since 1991. Meteorology atmospheric sciences, environmental sciences, and construction building technology were the three most frequent subject categories. Building and Environment, International Journal of Climatology, and Theoretical and Applied Climatology were the three most popular publishing journals. The USA and China were the two leading countries in UHI research, contributing 49.56% of the total articles. Chinese Academy of Science, Arizona State University, and China Meteorological Administration published the most UHI articles. Weng QH and Santamouris M were the two most prolific authors. Author keywords were classified into four major groups: (1) research methods and indicators, e.g., remote sensing, field measurement, and models; (2) generation factors, e.g., impervious urban surfaces, urban geometry, waste heat, vegetation, and pollutants; (3) environmental effects, e.g., urban climate, heat wave, ecology, and pollution; and (4) mitigation and adaption strategies, e.g., roof technology cooling, reflective cooling, vegetation cooling, and urban geometry cooling. A comparative analysis of popular issues revealed that UHI determination (intensity, heat source, supporting techniques) remains the central topic, whereas UHI impacts and mitigation strategies are becoming the popular issues that will receive increasing scientific attention in the future. Modeling will continue to be the leading research method, and remote sensing will be used more widely. Additionally, a combination of remote sensing and field measurements with models is expected.