A critical aspect in the design of tubular bridges is the fatigue performance of the structural joints. Economic viability depends on it. Lower fatigue strength for joints with thicker failing members was observed in welded joints typical to the bridge application. Different approaches to this phenomenon, called size effect, have been suggested, all based on the thickness correction for welded plate joints first proposed by Gurney. For the welded tubular joints, few studies on the size effects have been carried out; most of the existing investigations refer to geometries typical to petroleum industry offshore structures. In contrast to offshore structures, bridge structures have different absolute sizes and different member proportions (in particular lower chord radius to thickness ratios, γ). Tubular joints are far more complex than welded plate joints, multiple parameters are needed to describe the geometry (α, β, γ, τ, ζ) and there are several load scenarios. For these reasons, the fatigue behaviour analysis of such joints is a complex task. Current design recommendations combine the use of the structural (hot-spot) stress at the weld toe with a correction factor to take into account the wall thickness of the failing member. This approach oversimplifies the problem and can be very penalising, in particular for joints composed of thicker tubes, as is commonly the case for bridges. Furthermore, the truss member sizes that result from static design are likely to fall out of the validity range of current recommendations. This thesis focuses on a case of commonly used tubular joints: welded steel K-joint made out of circular hollow section (CHS). The main goals of this research are to understand the fatigue behaviour of as-welded CHS K-joints and to clarify the influences/effects of the different geometric parameters on their fatigue strength. In order to carry out a thorough study on the geometric size effects in CHS K-joints for bridges, fatigue tests were conducted for large-scale specimens with crack depth measurements and an advanced 3-D crack propagation model was developed. The first chapters of this thesis provide an introduction and a brief review of the main concepts in tubular joint fatigue and size effects on fatigue behaviour. The experimental tests of two tubular trusses under fatigue loading are then outlined. Crack growth in selected truss joints is monitored using the Alternating Current Potential Drop (ACPD) system. An advanced 3-D modelling of welded K-joint with surface crack is implemented using the boundary element method (BEM). A crack propagation model, based on Linear Elastic Fracture Mechanics (LEFM), is then developed using a step-wise incremental crack growth strategy. This model allows for fatigue strength and life estimations. Furthermore, it considers the influence of all geometric parameters that define CHS K-joints in a realistic way. The validation of the crack propagation model is made by comparisons with experimental data at different levels (i.e. member and joint strains and stresses, ACPD crack growth data). A parametric study is then carried out on joint geometries typical for a bridge application (low chord radius to thickness ratio) considering three basic load cases. Examples of results are shown and analysed on a "geometry cause"/"effect over the stress intensity factor and fatigue strength" basis. Parametric study results are then analysed, highlighting the case where the joint is proportionally scaled. The geometry correction factor, Y, is introduced as a function of the relative crack depth that is common to homothetic joints. The influence of the absolute size of the joint, also known as thickness effect, is determined for the three basic load cases. Parametric results are finally explored bringing to light the effect of non-proportional scaling. It is shown that size correction factors for fatigue strength can be expressed as a function of the non-dimensional geometrical parameters β, γ and τ, chord thickness, T, and different load cases. A new fatigue design method is proposed for welded (CHS) K-joints, based on LEFM and accounting for geometric size effects.