Methods for estimating spectral reflectance from XYZ colorimetry were evaluated using a range of different types of training datasets. The results show that when a measurement dataset with similar primary colorants (and therefore having similar reflectance curves) are used for training, the RMSE errors and metameric differences under different illuminants are the lowest. This study demonstrates that, a training data can be mapped to represent spectral data for a group of print data based on matching material components (spectral similarity) with the test data, and obtain spectral estimates with satisfactory spectral and colorimetric outcomes. The findings suggest that using polynomial bases or colorimetric weighted bases with least squares fit produced estimated reflectances with low metameric mismatches under different illuminants. For the two best performing spectral estimation methods their ability to predict tristimulus values were assessed with tristimulus calculated using the measured reflectances and a destination illuminant. Their performances were also compared to the colour predictions obtained from different CATs and MATs under varying lighting conditions. The results show that a spectral estimation method with specific training dataset can serve as a good alternative to predict XY Z under different illuminants with reduced metameric mismatch i.e. they can be used as a material adjustment transform. These results finally help in proposing a spectral estimation workflow that can be integrated into colour management such that it is simple to implement, fast in processing, spectrally accurate with low metameric mismatch.