Knowing how species respond to fire regimes is essential for ecologically sustainable management. This axiom raises two important questions: (1) what knowledge is the most important to develop and (2) to what extent can current research methods deliver that knowledge? We identify three areas of required knowledge: (i) a mechanistic understanding of species' responses to fire regimes; (ii) knowledge of how the spatial and temporal arrangement of fires influences the biota; and (iii) an understanding of interactions of fire regimes with other processes. We review the capacity of empirical research to address these knowledge gaps, and reveal many limitations. Manipulative experiments are limited by the number and scope of treatments that can be applied, natural experiments are limited by treatment availability and confounding factors, and longitudinal studies are difficult to maintain, particularly due to unplanned disturbance events. Simulation modelling is limited by the quality of the underlying empirical data and by uncertainty in how well model structure represents reality. Due to the constraints on large-scale, long-term research, the potential for management experiments to inform adaptive management is limited. Rather than simply recommending adaptive management, we define a research agenda to maximise the rate of learning in this difficult field. This includes measuring responses at a species level, building capacity to implement natural experiments, undertaking simulation modelling, and judicious application of experimental approaches. Developing ecologically sustainable fire management practices will require sustained research effort and a sophisticated research agenda based on carefully targeting appropriate methods to address critical management questions.