We present a theoretical and experimental study of erbium-doped fiber amplifiers in saturated operation, examining designs in which erbium doping is distributed throughout the core. We show the penalty due to reduced inversion near the core boundary is negligible, while the reduced amplifier length can improve performance. We examine the accuracy of compact models for such nonconfined doping and show that a widely-used compact model exhibits substantial errors, while the Giles-Desurvire (G-D) model (1991. IEEE J. Lightware Technol. 9, 271) is accurate, despite having been derived for confined erbium doping, Furthermore, we simplify the G-D model by eliminating the saturation parameters and argue that this can improve accuracy. Finally. we experimentally validate the simplified G-D model for a moderately confined fiber, comparing amplifier measurements with predictions based on measured fiber parameters; measured and modeled gain and noise figure are in excellent agreement, in the absence of adjustable parameters. (C) 2001 Academic Press.