Identifiability in statistical models refers to the ability to uniquely estimate model parameters from the observed data. It ensures that different parameter values do not produce the same probability distribution of the observed data, thus allowing for meaningful inference and prediction.