The paper analyzes conceptualizations in the science frameworks in three large-scale assessments, Trends in Mathematics and Science Study (TIMSS), Programme for International Student Assessment (PISA), and National Assessment of Educational Progress (NAEP). The assessments have a shared history, but have developed different conceptualizations. The paper asks "how" and "why" the frameworks are different and seeks answers by studying their development. The methodology is document analysis by, first, tracing developments within each assessment, next, comparing developments and conceptualization across the assessments, and, last, relating the frameworks to trends of developments in educational theory. The outcome of the analysis provides a complex picture with the assessments following their own lines of development but with influence from trends in assessment and educational theory. Five main conceptualizations are found to have existed over time, with different definition of "scientific behavior" and explanations to the relationship between knowledge and behavior. The frameworks have moved toward more elaborated "explanations" of the science domain, providing assessors with better support for operationalizing learning objectives. Currently, the assessments are faced with a challenge of adapting to the "practice turn" in science studies and learning science and thereby accounting for scientific behavior as a "community practice." The paper concludes with suggestions for how frameworks may be improved to achieve this aim.