Global miRNA expression arrays were used for analysis of 836 miRNAs in formalin-fixed paraffin-embedded samples from 21 tongue cancer patients and 8 controls. Samples had been stored for one to eleven years. Results separated tumour samples from controls, however, the largest variation was correlated to sample storage time, detectable already after one year. With the use of a linear regression model we could adjust for the storage-dependent effect, leading to the identification of 54 differentially expressed miRNAs in tongue cancer, compared to 16 when using standard normalization, including up-regulation of a novel miRNA, miR-424.