Milling properties are an important component of the commercial value of durum wheat (Triticum turgidum L. subsp. durum (Desf.)Husn.), and are thus important considerations in breeding. However, measurement of milling-related traits is costly, so evaluation is usually deferred to later filial generations when genetic diversity may be lost. The objective of this work was to identify quantitative trait loci (QTL) that could potentially be used for marker-assisted selection of milling traits. A diversity panel of 87 lines was assessed for semolina yield, residual flour, total mill extraction, semolina ash, and semolina agtron colour. The diversity panel sampled lines from 13 countries and was grown in field trials arranged as a 10 x 10 lattice design with two replications grown at Regina, SK and Vauxhall, AB in 2001 and 2002. The lines were genotyped with the wheat Illumina 90K iSelect assay. For each trait, SAS PROC MIXED was performed on each year with blocking based on the week the material was milled. The Azzalini-Cox test was performed to identify lines that had crossover interactions. Milling data were also adjusted using test weight as a covariate because of the observed negative correlation between test weight and several milling traits in this population. For total extraction, each of the three lines Bonaerense Valverde, Borli and Tresor had noticeable differences in relative ranking among the four environments, especially between Regina 2002 and each of the other three environments. Dropping these three lines consistently improved the Pearson’s correlations among all pairwise combinations of environments, with the greatest improvements between Regina 2002 and each of the other three environments. When these three lines were dropped from the marker association analysis with G-Model (significant P<10-6), three consistent markers on chromosomes 1B, 3A and 7A were identified in all environments whereas when these three lines were included, the markers were not identified in any environments. Adjustment for test weight and dropping two lines that continued to show GxE crossover interaction further improved consistency, with the SNPs on chromosomes 3A and 7A identified in all four environments, indicating they were independent of test weight. The 1B SNP was no longer significant, suggesting the region was associated with test weight. These results emphasize the importance of thorough analysis of phenotypic data in maximizing the information that can be extracted from association analyses. This is especially critical for traits such as milling, wherein replication is limited by low throughput of phenotyping.