We investigate the crystallization history of an early Paleozoic granitic suite within the Arden pluton, Wilmington Complex, Pennsylvania-Delaware Piedmont, and its implications for granite petrogenesis. The pluton intruded into granulite facies gneisses at a depth of 17-25 km, and magmatic textures and mineral and rock compositions are generally well preserved. The granitic suite is characterized geochemically by a wide range of rock compositions (e.g., 54-76 wt% SiO 2 ; 0.7-5.7 wt% K 2 O); strong linear correlation among all elements except Ba, verified by principal components analysis; large apparently random scatter in Ba concentration; limited variation in mineral composition among rocks ranging from 58-73 wt% SiO 2 ; and no apparent spatial pattern to rock chemical variation within the pluton.We explain these factors by a model of in situ crystallization within a solidification zone (or boundary layer). We derive equations that integrate both major-and trace-element variations in a quantitative model of the migration and crystallization of residual liquid within a mush of relatively earlier formed crystals. Barium is the key element that traces the history of melt migration, because it changed behavior from incompatible to compatible with the onset of K-feldspar crystallization from the residual granitic liquid. The model explains the large range in rock chemistry as well as the linear correlations among elements in terms of the proportions of earlier formed crystals, and minerals that crystallized from the residual melt. The apparent scatter of Ba concentration in the rocks resulted from the variable depletion of Ba in the residual liquid as it crystallized and migrated through the crystal mush, and requires that trapped liquid did not equilibrate in trace-element composition on a scale of meters or more within the crystal mush.From our equations, we derive values for parameters that describe the proportion of melt that remained after crystallization of earlier formed minerals, and the proportion of residual melt that ultimately left the rock. Net melt loss for most rocks is ≥50%, suggesting that melt was not only migrating locally, but was being transported from the level of the pluton now exposed. This result may be more compatible with gravity-influenced processes near the base of a growing crystal pile, rather than with sidewall crystallization. Nearly uniform compositions and proportions of the relatively earlier formed crystals could be explained by efficient compositional convection between the granitic magma reservoir and residual melt circulating through the crystal mush. There is a roughly concentric spatial pattern to the model parameters for melt proportions that suggests an outer zone of high initial crystallinity and little melt loss near the present margin of the pluton, an interior zone with large losses of residual melt that is consistent with high melt fluxes and compositional convection, Srogi, L., and Lutz, T. M., 1997, Chemical variation in plutonic rocks caused by residual melt migrat...