A common interpretation of existing subjective well-being research is that long-term levels of well-being are almost completely stable. However, few studies have estimated stability and change using appropriate statistical models that can precisely address this question. The STARTS model (Kenny & Zautra, 2001) was used to analyze life satisfaction data from two nationally representative panel studies. Results show that 34-38% of the variance in observed scores is trait variance that does not change. An additional 29-34% can be accounted for by an autoregressive trait that is only moderately stable over time. Thus, although life satisfaction is moderately stable over long periods of time, there is also an appreciable degree of instability that might depend on contextual circumstances.