Scholars have identified mathematics gender gaps favoring males as early as kindergarten or first grade, particularly at the top of the achievement distribution (Penner & Paret, 2008; Rathbun, West & Germino-Hausken, 2004; Robinson & Lubienski, 2011). These relatively small achievement disparities precede larger differences in students' career choices. For example, men recently earned 82% of engineering bachelor's degrees, while women earned only 18% (Dey & Hill, 2007). Women's under-representation in math-related careers both limits the pool of talented people contributing to those fields and leaves disproportionate numbers of women in lower-paying occupations. In examining the possible origins of these early math gender gaps, previous researchers looked inside mathematics classrooms and found that teachers tended to hold higher expectations of their male students and to view mathematics as a male domain (Li, 1999). Yet, in contrast to this previous work, recent, large-scale studies suggest that teachers actually rate the performance of girls more favorably than the performance of males (e.g., Fryer & Levitt, 2010; Robinson & Lubienski, 2011). Given gender disparities in mathematics-related careers, the new findings seem to be promising news if teachers' positive assessments help level the playing field for future generations of women in STEM careers. However, these initial estimates of teachers' female bias may be misleading, confounding achievement with behavior and learning approaches. Indeed, prior research has revealed that girls tend to exhibit more on-task behavior and positive approaches to learning behavior in schools (Forgasz & Leder, 2001; Ready, LoGerfo, Lee & Burkam, 2005). Hence, teachers might conflate "good girl" behavior with mathematics proficiency. This study untangles these issues, examining whether teachers in a national sample rate boys' math proficiency higher than that of girls when boys and girls behave similarly, have similar approaches to learning, and have the same past and current test scores. This study also examines whether teachers' tendency to rate boys or girls higher is causally linked to the widening gender gap in mathematics in early elementary school. In prior research, mathematics achievement gaps favoring males were found to widen during early elementary school; however, teachers tended to rate girls' mathematics proficiency higher than that of boys with similar mathematics test scores (Robinson & Lubienski, 2011). This research builds upon this prior work by examining the following two research questions: (1) Do teachers still rate the mathematics proficiency of girls higher when boys and girls are equated in terms of demographics, prior achievement, behavior, and teacher-reported approaches to learning? (Study 1); and (2) If teachers do have a tendency to rate observationally-similar boys and girls differently, do these differential ratings have an effect on the development of the mathematics gender gap in elementary school (Study 2)? This research uses the Early Childhood Longitudinal Study, Kindergarten Class of 1998-99 (ECLS-K), which is nationally representative of the kindergarten class of 1998-99 when the NCES-provided sampling weights are used. As Robinson and Lubienski (2011) demonstrated, the math gender gap develops early--in the first few years of formal schooling, growing from nonexistent in the fall of kindergarten to a male advantage of about 0.25 standard deviations by third grade. Study 1 demonstrates that teachers rate the math skills of girls lower than those of observationally similar boys. That is, conditioning on math achievement histories, behavior, approaches to learning, race, age, SES, and even looking at boys and girls with the same teachers, girls' skills are rated to be more than one-tenth of a standard deviation lower than boys. This pattern is consistent throughout elementary school. Lamentably, even when conditioning on "current" math achievement, girls are still rated lower (as shown in Figure 1). There is no evidence of similar ratings disadvantage for black or Hispanic students; and there is no evidence that girls are rated higher in reading. Thus, this teacher underrating phenomenon is unique to girls and math performance. Study 2 demonstrates that girls lose ground in math to boys in every period examined (from the spring of kindergarten through fifth grade), consistent with recent studies (Fryer & Levitt, 2010; Robinson & Lubienski, 2011). However, when the authors account for the effects of teachers' expectancies, they find that girls lose far less ground. Their analyses tested the instruments used (i.e., they tested if prior teacher ratings were correlated with conditional achievement gains in a way other than through teacher ratings), and they found no evidence to suggest they were invalid. Overall, the results suggest if teachers did not believe that boys had higher math proficiency than similar girls, then girls would lose about 40-75% less ground in math achievement in each period examined. Raising awareness of--and hopefully, reducing--the tendency for teachers to rate males higher in math may thus go a long way to close the gender achievement gap in math. (Contains 4 figures, 2 tables and 2 footnotes.