The invention relates essentially to a method of detecting a fall of a person, characterized in that it comprises the following steps:- obtaining a signature St at an instant t, representing the distribution over a time window Δt of the kinetic energy, measured in three dimensions by an accelerometric sensor,- normalising the said signature,- comparing it with a reference signature obtained for a representative sample of falls, by means of similarity measurements,- analysing the result of the comparison, and- possibly, triggering a signal warning of a fall.A similar method based on cardiovascular parameters can be combined with the above processing to improve its reliability.