This paper compares eight background correction methods for two-colour microarrays to determine the best approach for differential expression analysis. The methods include standard background subtraction, Kooperberg, Edwards, normexp, vsn, morph, normexp+offset, and no background. The study uses three data sets: a spike experiment, a mixture experiment, and a quality control study. The results show that methods which stabilize the variance of log-ratios perform best. The normexp+offset method has the lowest false discovery rate, followed by morph and vsn. These methods produce strictly positive corrected intensities and avoid missing values. The standard method, which subtracts local background estimates, produces negative intensities and high variability in low intensity log-ratios. The study concludes that background correction methods should be chosen based on the trade-off between bias and precision. The normexp+offset and morph methods are recommended for their ability to stabilize variance and improve differential expression analysis. The study also highlights the importance of using appropriate background correction methods to avoid missing values and improve the accuracy of differential expression analysis. The results suggest that the best background correction methods are those that stabilize the variance as a function of intensity, such as normexp+offset, morph, and vsn. The study emphasizes the need for careful selection of background correction methods to ensure accurate and reliable differential expression analysis.This paper compares eight background correction methods for two-colour microarrays to determine the best approach for differential expression analysis. The methods include standard background subtraction, Kooperberg, Edwards, normexp, vsn, morph, normexp+offset, and no background. The study uses three data sets: a spike experiment, a mixture experiment, and a quality control study. The results show that methods which stabilize the variance of log-ratios perform best. The normexp+offset method has the lowest false discovery rate, followed by morph and vsn. These methods produce strictly positive corrected intensities and avoid missing values. The standard method, which subtracts local background estimates, produces negative intensities and high variability in low intensity log-ratios. The study concludes that background correction methods should be chosen based on the trade-off between bias and precision. The normexp+offset and morph methods are recommended for their ability to stabilize variance and improve differential expression analysis. The study also highlights the importance of using appropriate background correction methods to avoid missing values and improve the accuracy of differential expression analysis. The results suggest that the best background correction methods are those that stabilize the variance as a function of intensity, such as normexp+offset, morph, and vsn. The study emphasizes the need for careful selection of background correction methods to ensure accurate and reliable differential expression analysis.