INTRODUCTION: The global burden of breast cancer is increasing, however secondary prevention via population-based mammographic screening has proven effective in reducing mortality in high income countries. Concomitant with screening are various drawbacks including over diagnosis and false positives, hence new methods for improving screening performance are vital if the balance of “benefit versus harm” is to be improved. The emergence of automated tools for large-scale objective image analysis offers new possibilities. The overall aim of this thesis was to look at three ways that such tools might be used for improving breast screening performance. OBJECTIVES: (1) To assess whether automatically estimated volumetric breast measurements could be used as a proxy for BMI (a confounder in the association between breast density and breast cancer risk) in screening settings where BMI data are not available. (2) To use a novel, automatically estimated measurement of left breast versus right breast fluctuating asymmetry (FA), in breast volume and mammographic density, to determine whether FA is associated with cancer detection at screening or the occurrence of cancer in the interval between screens (i.e. interval cancers). (3) To assess whether variations in, objectively measured, mammographic compression force, pressure and paddle tilt are associated with screening performance. METHODS: For objective (1) data from a previously-conducted UK case-control study (414 cases/685 controls) and a Norwegian cohort study (657 cases/61,059 non-cases) were pooled using fixed-effect models (Study I). For objectives (2) and (3) four studies (II-V) were designed, requiring data collation and image analysis of over 90,000 screens (from which 904 cancer cases were detected) at a UK population screening programme. Cross-sectional designs were used for assessing the association between relevant exposures and breast cancer detection at screening, and nested case-control designs for assessing the association of exposures with the occurrence of an interval cancer. RESULTS: 1) Study I confirmed that, in screening age women, non-dense breast volume is strongly and positively correlated with BMI (r=0.74, p<0.0001) and it showed empirically for the first time, that using breast volume estimates in place of BMI leads to minimal difference in the association between % mammographic density and breast cancer risk (pooled RR 1.51(95%CI 1.41-1.61) in both cases). 2) Studies II and III showed that mammographic density FA is common and that women with highest FA were more likely to be diagnosed with cancer at screening (OR=1.26 (95% CI 1.07-2.27) for top versus bottom third of the distribution; P for linear trend=0.012). Similarly, women in the top third of FA were at higher risk of interval cancer (OR=1.68 (95% CI 0.97-2.92)). 3) Studies IV and V found that breast compression thickness decreased with increased compression force and pressure, but that increasing pressure above ~15kPa resulted in minimal further reduction. Compression pressure was negatively associated with detection of cancer at screening (OR=0.74 (95% CI 0.60- 0.92) for the top versus bottom third of the pressure distribution). CONCLUSIONS: These findings showed that automated mammogram analysis tools have the potential to be used in novel ways in breast cancer screening and, more widely, in breast cancer risk assessment even in high-volume screening settings where it is not feasible to routinely collect BMI data. FA, a novel automated measurement, may help to identify women at higher breast cancer risk and those more likely to have an interval cancer. Finally, my thesis challenges the view that using ‘as much force as tolerated’ during mammography is the best strategy and suggests that there are more subtle associations between breast compression technique and screening outcomes.