From a practical point of view, some measures to reduce the thickness effect backed by a reasonable criterion are required for fabricating structures with heavy section plates. In this study, the thickness effect was investigated by systematic experiments on welded steel joints with thicknesses ranging from 10 to 80 mm. Cruciform joints and T-joints with improved weld by overall profiling or toe-grinding were tested under pulsating tension and under pulsating bending, respectively. These experimental results were analyzed together with the previous results of as-welded joints. As a result, it was concluded that the thickness effect exponents for various conditions may be classified into three categories according to the combination of joint type and loading mode. As-welded joints under bending stress have the steepest thickness effect exponent of −1/3, while as-welded joints under tension with an exponent of −1/5 is milder in thickness effect than that specified in the existing codes. If the weld profile is improved by grinding, the thickness effect becomes much milder to an exponent of −1/10. The as-weld joints with constant-sized attachments also have an exponent of −1/10. Furthermore, thickness effect dependence on fatigue life and thickness effect under random stress were investigated. Based on these results, this study proposes a new evaluation criterion for design purposes.