Fatigue life assessment of steel highway bridges is crucial to maintain their safety. Researches are required to quantify uncertainties in loading, resistance, fatigue life prediction and improvement techniques. This study calibrates some fatigue code parameters and models fatigue loading probabilistically using truck data in Canada. Fatigue life improvement by weld toe grinding is assessed experimentally and numerically. Finally, the remaining fatigue life using reliability-based method for cruciform bridge welded detail is predicted. Truck data screening is vital to ensure its quality in calibration and load history prediction. Three different fatigue truck factors and equivalent number of cycles are proposed, using single and dual slope curves, for three bridge span ranges. Span length and fatigue category affected the calibration in short spans and dual slope curves. Using Ontario and Quebec data, probabilistic distributions for stress-range histograms for bridge spans and configurations are developed. Weld toe grinding improved the fatigue life by 60% for category E joint compared to the non-ground specimen. Still, weld imperfections might adverse grinding effect. In deterministic fatigue life prediction, energy-based approach is used for crack initiation, while linear elastic fracture mechanics is used for crack propagation. An initial spherical flaw size of 0.1 mm with no residual stresses leads to the most precise prediction of fatigue resistance for test specimens. For the probabilistic fatigue life prediction of cruciform detail of 6.4 mm fillet weld size, the fatigue life of the concave weld profiles are usually larger than the convex ones. The study developed a method to convert the probabilistic stress range into number of cycles for remaining fatigue life reliability of steel detail. The generated reliability illustrates the fatigue repair and inspection management.