Background/Context: We lack strong and consistent information about which measures of knowledge matter most for good teaching and student learning, and what are trajectories of improvement for novice teachers. Research Questions: We explore the level, variation, and change in teacher knowledge and instruction in the first two years of teaching, the relationship between Mathematical Knowledge for Teaching (MKT) and more distal measures such as certification. Sample: We studied 45 middle school math teachers in their first two years of teaching, in 11 districts of varying size and urban status in two southeastern and two mid-Atlantic states. Research Design and Analysis: This is a longitudinal (two-year) study of natural variation, which includes descriptive, correlational, individual growth curve, and regression analyses. Data Collection: Based on multiple administrations of survey data, MKT assessments, and classroom observations using the Instructional Quality Assessment (IQA), we developed measures of (a) the rigor of lesson activities and classroom discussion, (b) the quality of classroom discussion, (c) the relative emphasis on procedural versus higher-order cognitive demands, (d) the proportion of time spent on basic versus advanced math topics, and (e) the number of topics covered, or instructional breadth. Findings: Key findings are as follows: (1) many beginning math teachers in our sample had neither a degree in math nor substantial coursework in math; (2) teachers generally had low MKT scores, a balanced approach to emphasizing cognitive demands, low levels of discussion quality, and substantial across-teacher variation in topic coverage; (3) teachers improved in some but not all measures of instructional quality; (4) there were no direct relationships between MKT and instructional quality; (5) we found little evidence that MKT is a better predictor of instructional quality than distal measures, but we did find suggestive evidence that MKT may help to explain their predictive power; (6) we found suggestive evidence that taking more advanced math courses predicts desirable teaching practices; and (7) the number of weeks of student teaching in math was consistently related to more rigorous instruction and less emphasis on basic topics. Conclusions: These results have implications for shaping teacher preparation programs, teacher in-service professional development, and certification policies, as well as how we study new teachers and calibrate our expectations for improvement in novice teachers.