In targeted therapy, patient tumors are analyzed for aberrant activations of core cancer pathways, monitored based on biomarker expression, to ensure efficient treatment. Thus, diagnosis and therapeutic decisions are often based on the status of biomarkers determined by immunohistochemistry in combination with other clinical parameters. Standard evaluation of cancer specimen by immunohistochemistry is frequently impeded by its dependence on subjective interpretation, showing considerable intra- and inter-observer variability. To make treatment decisions more reliable, automated image analysis is an attractive possibility to reproducibly quantify biomarker expression in patient tissue samples. We tested whether image analysis could detect subtle differences in protein expression levels. Gene dosage effects generate well-graded expression patterns for most gene-products, which vary by a factor of two between wildtype and haploinsufficient cells lacking one allele. We used conditional mouse models with deletion of the transcription factors Stat5ab in the liver as well Junb deletion in a T-cell lymphoma model. We quantified the expression of total or activated STAT5AB or JUNB protein in normal (Stat5ab+/+ or JunB+/+), hemizygous (Stat5ab+/Δ or JunB+/Δ) or knockout (Stat5abΔ/Δ or JunBΔ/Δ) settings. Image analysis was able to accurately detect hemizygosity at the protein level. Moreover, nuclear signals were distinguished from cytoplasmic expression and translocation of the transcription factors from the cytoplasm to the nucleus was reliably detected and quantified using image analysis. We demonstrate that image analysis supported pathologists to score nuclear STAT5AB expression levels in immunohistologically stained human hepatocellular patient samples and decreased inter-observer variability.