We analysed chromosome 16q in 106 breast cancers using tiling-path array-comparative genomic hybridization (aCGH). About 80% of ductal cancers (IDCs) and all lobular cancers (ILCs) lost at least part of 16q. Grade I (GI) IDCs and ILCs often lost the whole chromosome arm. Grade II (GII) and grade III (GIII) IDCs showed less frequent whole-arm loss, but often had complex changes, typically small regions of gain together with larger regions of loss. The boundaries of gains/losses tended to cluster, common sites being 54.5-55.5 Mb and 57.4-58.8 Mb. Overall, the peak frequency of loss (83% cancers) occurred at 61.9-62.9 Mb. We also found several 'minimal' regions of loss/gain. However, no mutations in candidate genes (TRADD, CDH5, CDH8 and CDH11) were detected. Cluster analysis based on copy number changes identified a large group of cancers that had lost most of 16q, and two smaller groups (one with few changes, one with a tendency to show copy number gain). Although all morphological types occurred in each cluster group, IDCs (especially GII/GIII) were relatively overrepresented in the smaller groups. Cluster groups were not independently associated with survival. Use of tiling-path aCGH prompted re-evaluation of the hypothetical pathways of breast carcinogenesis. ILCs have the simplest changes on 16q and probably diverge from the IDC lineage close to the stage of 16q loss. Higher-grade IDCs probably develop from low-grade lesions in most cases, but there remains evidence that some GII/GIII IDCs arise without a GI precursor.