It is contended that safety-arguments can be improved by being made “systematically self-reflecting” (SSR). A safety argument should systematically address its own quality at each point in its argumentation. This can be done by addressing specific questions at each argument node, without changing the structure of the argument. The approach has been applied to software safety-cases expressed using the Goal Structuring Notation (GSN), and lessons are drawn from this industrial experience. A comparison is made between SSR-arguments and Kelly's criteria for assessing assurance-cases, the 4+1 principles for software safety-cases and the confidence-arguments of Hawkins et al.