The　electrical　activity　of　the　heart　shows　complex　dynamical　behavior.　This　paper　presents　a　nonlinear　local　projective　noise　reduction　filter　for　electrocardiograms.　In　order　to　automatically　detect　the　best　neighborhood　size　of　the　projection,　recurrence　plots　are　used.　The　minimum　gives　a　clear　indication　of　the　best　size　of　the　neighborhood　in　the　embedding　space.　We　also　use　the　box-assisted　algorithm　to　efficiently　find　nearest　neighbors　in　m-dimensional　space.　The　method　is　successfully　applied　to　ECG　signals.　©　2008　IEEE.