Brosnan and Pearlstein showed that the zero locus of an admissible normal function on an algebraic variety is itself algebraic. I will explain a different proof of this result, based on "Néron models" for intermediate Jacobians.