The intensities from genotyping array data can be used to detect copy number variants (CNVs) but a high level of noise in the data and overlap between different copy-number intensity distributions produces unreliable calls, particularly when only a few probes are covered by the CNV. We present a novel pipeline (CamCNV) with a series of steps to reduce noise and detect more reliably CNVs covering as few as three probes. The pipeline aims to detect rare CNVs (below 1% frequency) for association tests in large cohorts. The method uses the information from all samples to convert intensities to z-scores, thus adjusting for variance between probes. We tested the sensitivity of our pipeline by looking for known CNVs from the 1000 Genomes Project in our genotyping of 1000 Genomes samples. We also compared the CNV calls for 1661 pairs of genotyped replicate samples. At the chosen mean z-score cut-off, sensitivity to detect the 1000 Genomes CNVs was approximately 85% for deletions and 65% for duplications. From the replicates, we estimate the false discovery rate is controlled at ∼10% for deletions (falling to below 3% with more than five probes) and ∼28% for duplications. The pipeline demonstrates improved sensitivity when compared to calling with PennCNV, particularly for short deletions covering only a few probes. For each called CNV, the mean z-score is a useful metric for controlling the false discovery rate. © 2020 The Authors. Genetic Epidemiology published by Wiley Periodicals LLC.