Intracerebral hemorrhage (ICH) growth predicts mortality and functional outcome. We hypothesized that irregular hematoma shape and density heterogeneity, reflecting active, multifocal bleeding or a variable bleeding time course, would predict ICH growth. Three raters examined baseline sub-3-hour CT brain scans of 90 patients in the placebo arm of a Phase IIb trial of recombinant activated Factor VII in ICH. Each rater, blinded to growth data, independently applied novel 5-point categorical scales of density and shape to randomly presented baseline CT images of ICH. Density and shape were defined as either homogeneous/regular (Category 1 to 2) or heterogeneous/irregular (Category 3 to 5). Within- and between-rater reliability was determined for these scales. Growth was assessed as a continuous variable and using 3 binary definitions: (1) any ICH growth; (2) >or=33% or >or=12.5 mL ICH growth; and (3) radial growth >1 mm between baseline and 24-hour CT scan. Patients were divided into tertiles of baseline ICH volume: "small" (0 to 10 mL), "medium" (10 to 25 mL), and "large" (25 to 106 mL). Inter- and intrarater agreements for the novel scales exceeded 85% (+/-1 category). Median growth was significantly higher in the large-volume group compared with the small group (P<0.001) and in heterogeneous compared with homogeneous ICH (P=0.008). Median growth trended higher in irregular ICHs compared with regular ICHs (P=0.084). Small ICHs were more regularly shaped (43%) than medium (17%) and large (3%) ICHs (P<0.001). Small ICHs were more homogeneous (73%) compared with medium (37%) and large (17%) ICHs (P<0.001). Adjusting for baseline ICH volume and time to scan, density heterogeneity, but not shape irregularity, independently predicted ICH growth (P=0.046) on a continuous growth scale. Large ICHs were significantly more irregular in shape, heterogeneous in density, and had greater growth. Density heterogeneity independently predicted ICH growth using some definitions.