This paper uses firm-level survey data matched with official tax data to estimate the true sales of formal firms in Mongolia using a structural equation modeling with latent variable for which “multiple-indicators and multiple-causes (MIMIC)” exist. Underreporting in sales to tax office is calculated by comparing the predicted true sales from the MIMIC model with tax office sales. A direct approach and an indirect approach are also used to estimate the size of underreporting in sales. A direct approach is based on firm-level perceptions of underreporting by the “typical establishment” in the survey firm’s area of activity and gives a reasonable estimate of mean % of underreporting. An indirect approach is based on a comparison of the survey sales with tax office sales and suggests low levels of underreporting to tax office due to measurement error and underreporting in the survey sales. The MIMIC approach does not rely on perception data, explicitly controls for measurement error, and allows for underreporting in both tax and survey data. Hence we argue that MIMIC approach gives the most accurate estimates of underreporting in sales. According to the MIMIC approach, 37.7% of the aggregate sales are underreported to tax office and the firm-level mean % of underreporting to tax office is 38.7%. Underreporting to tax office varies across firms. Conditional on true sales, underreporting in sales to tax office increases with firm size and the average skill level of workers but decreases with manager experience, and firms outside of Ulaanbaatar underreport more. Though the proxies of corruption burden and constrained access to finance don’t affect conditional (on true sales) underreporting to tax office, they are significantly negatively associated with unconditional underreporting in sales. The MIMIC approach shows that significant underreporting also occurs in the survey and it increases with firm size. This has important implications for economic studies using firm-level survey data.