The high temporal variability of the soil-to-atmosphere CO 2 flux (soil respiration, R S ) has been studied at hourly to multiannual time scales but remains less well understood than R S spatial variability. How R S fluxes vary and are autocorrelated at various time lags has practical implications for sampling and more fundamentally for our understanding of its abiotic and biotic underlying mechanisms. We examined the variability, correlation, and sampling requirements of R S over a wide range of temporal scales in a temperate deciduous forest in eastern Maryland, USA, using both automated (temporally continuous, N = 30,036 over 10 months) and survey (spatially diverse, temporally sparse, N = 1,912 over 17 months) data. Data from a global R S database were also used to examine interannual variability in comparable forests. The coefficient of variability of successive measurements generally varied from the minute (median coefficient of variation 16%) to hourly and daily (11-12%) time scales. Successive R S values measured at a given collar exhibited a strong hour-to-hour correlation (r = 0.931) and a moderate correlation at a 2-hr lag (0.289); day-to-day (i.e., 24 hr lag) hourly observations were uncorrelated. Daily R S means were well correlated at a 1-day lag (r = 0.856) but not at any further time lag. In a linear mixed-effects model predicting R S , soil temperature and moisture exerted consistently strong effects regardless of time scale, and model coefficient of variability was generally high (>80%). These results provide new opportunities to explore the drivers and variability of R S fluxes, quantify sampling requirements, and improve error propagation.Plain Language Summary Soils give off carbon dioxide, generated by microbes and plant roots, to the atmosphere. How this "soil respiration" (R S ) varies in time, as one measures at minute, hourly, daily, or longer time scales, is related to the processes driving it and has implications for how we estimate this flow of carbon across space and time. We measured R S in a coastal deciduous forest in Maryland, USA, and found that the variability of R S -how much it changed between successive measurementsvaried at the different time scales. R S values quickly became increasingly random (i.e., uncorrelated with each other) as one measured repeatedly over time. These results help us understand the factors driving this large flow of carbon to the atmosphere and improve our ability to estimate R S at times and places not directly measured.