We assess the biases induced by measurement error in a study of correlations in educational attainment across two and three generations in the United States using linked data spanning 1940–2015. Although multigenerational regressions show an economically meaningful “grandparent effect,” we find that this effect is overstated and likely not statistically or economically significant after we correct for measurement error. However, that same measurement error means that inferences derived using only two generations of data understate persistence in educational attainment by roughly 20%. We find an 18% decline in the intergenerational correlation of educational attainment over the twentieth century.