The detection of copy number variants (CNV) by array-based platforms provides valuable insight into understanding human diversity. However, suboptimal study design and data processing negatively affect CNV assessment. We quantitatively evaluate their impact when short-sequence oligonucleotide arrays are applied (Affymetrix Genome-Wide Human SNP Array 6.0) by evaluating 42 HapMap samples for CNV detection. Several processing and segmentation strategies are implemented, and results are compared to CNV assessment obtained using an oligonucleotide array CGH platform designed to query CNVs at high resolution (Agilent). We quantitatively demonstrate that different reference models (e.g. single versus pooled sample reference) used to detect CNVs are a major source of inter-platform discrepancy (up to 30%) and that CNVs residing within segmental duplication regions (higher reference copy number) are significantly harder to detect (P < 0.0001). After adjusting Affymetrix data to mimic the Agilent experimental design (reference sample effect), we applied several common segmentation approaches and evaluated differential sensitivity and specificity for CNV detection, ranging 39-77% and 86-100% for non-segmental duplication regions, respectively, and 18-55% and 39-77% for segmental duplications. Our results are relevant to any array-based CNV study and provide guidelines to optimize performance based on study-specific objectives.