Array comparative genomic hybridization (aCGH) can detect copy number variation (CNV) across the genome. Five current Hidden Markov Model (HMM) software systems for estimating copy number variation with aCGH data were compared. These comparisons were in terms of their effectiveness for identifying CNVs in simulated data based on the ratio of signal intensities. There was significant variability in the error rates. The system that adjusted for outliers in the model, the Robust Hidden Markov Model (HMM-R), appeared to have the best performance. The emission density function of the HMM is a mixture of two normal densities, in which one component represents usable aCGH data and the other represents outliers. HMM-R correctly classified 99.8% of normal states, 84.5% of CNV gains, and 90.2% of CNV losses. That is, error rates with regard to gains and losses were appreciable even with the best software. The HMM-R method demonstrated higher sensitivity and lower false discovery rates than the commonly used procedure. While the accuracy rates of HMM software has improved, there is substantial room for further improvement.