Although the question of animal welfare has been an important source of concern in the scientific community for several decades, many aspects are still under debate. On-farm assessments have to be rapid, acceptable to farmers and safe for both the assessors and animals. They are thus very demanding, with multiple decisions to make, such as the choice of appropriate indicators, sampling methods and scoring. Research has moved from resource-based to animal-based criteria, which reflects the subjective welfare state of an animal rather than relying upon external indices. In the present review, we describe two major (i.e., the most frequently/recently tested or disseminated) protocols: one in low-/middle-income countries, and the other in high-income countries, for on-farm assessments of horses, using animal-based resources; we evaluate their strengths and limitations, and then we compare their results with those obtained by various other studies. We propose lines of improvement, particularly in view of public dissemination, and offer suggestions for further refinement or new protocols. We emphasize the high risks of putting the cart before the horse, i.e., proposing protocols that rely upon indicators and sampling methods that need to be refined, as this could lead to under-evaluation (or less likely over-evaluation) of current welfare problems. Because welfare is a subjective experience, the true representation of an individual's actual welfare status has to be evaluated by using objective assessment tools (that are validated and have a scientific basis) used by well-trained observers.