Recents studies have shown that the use of improvement techniques on welds allows an increase in the level of admissible fatigue strength (even more than 100%). The aim of this study was to collect and validate literature data, create a data base containing the joint characteristics and fatigue results, make a statistical analysis of the data in order to quantify the effect of parameters influencing the fatigue strength and propose new S-N curves which are compared to those given in Eurocode 3. Four improvement techniques (grinding, TIG dressing, hammer peening, shot peening) and four joint types (butt, T joints, cruciform and longitudinal joints) were taken into account. Joint thicknesses less than 25 mm loaded in air with a stress ratio R between 0 and 0.1 were selected. Three classes of yield strength for the base metal were considered: <100, 100-600, >600. All S-N curves were above those of as-welded assemblies. The best results were obtained with hammer peening. The larger increase in the fatigue strength due to the use of improvement techniques was due to the occurrence of an initiation phase in addition to the crack propagation phase. During the initiation phase, the extension of existing crack-like defects is slowed down or even stopped. The duration of this phase increases with the total fatigue life.