The first stage in the analysis of cDNA microarray data is estimation of the level of expression of each gene, from laser scans of the hybridised microarrays. Combining multiple laser scans can be formulated as multivariate functional regression through the origin. Maximum likelihood estimation fails, but many alternative estimators exist, one of which is to maximise the likelihood of a Gaussian structural regression model. We show by simulation that, surprisingly, this estimator is efficient even though the distribution of gene expression values is far from Gaussian.