We present a novel Hidden Markov Model for detecting copy number variations (CNV) from genotyping arrays. Our model is a novel application of HMM to inferring CNVs from genotyping arrays: it assumes a continuous-time framework and is informed by prior findings from previously analyzed real data. This framework is also more realistic than discrete-time models which are currently used since it does not assume that CNV breakpoints occur at the genotyped loci. We show how to estimate the model parameters using a training data of normal samples whose CNV regions have been confirmed, and present results from applying the model to a set of HapMap samples containing aberrant SNPs.