Previous studies have shown that people make letter-reversal errors in identifying strings of letters in peripheral vision. These errors contribute to a rapid fall-off of letter-recognition performance away from fixation. This study tests the hypothesis that these errors are due to decreased precision (increased "noise") in the coding of letter position in the periphery. To test our hypothesis, 4 observers identified pairs of adjacent letters that were presented within 8 letter positions left and right of fixation, and judged the relative position of these two letters (which one was to the left). Letters of each pair were presented sequentially (each for 50 ms) and separated by a mask (50 ms). Character sizes were 0.3, 0.5 and 0.8 deg. Trials in which observers did not correctly identify at least one letter (regardless of letter position) were excluded in our analysis. To estimate the noise in coding letter position, we used a model in which the encoded position of each letter ("local sign") is Gaussian distributed with a mean equal to the true retinal eccentricity and standard deviation (SD) that depends on eccentricity. The SD of the noise can then be derived from the percent-correct data in judging the relative position for each pair of letters. Plots of percent-correct as a function of letter position showed an inverted U-shaped function for both the tasks of letter identification and relative letter position judgment. When expressed in units of letter spaces, the derived SD of the position noise increases with eccentricity, and is independent of character size. Across observers and character sizes, the median SD is approximately 0.5 letter spaces in the fovea, and increases to approximately 0.96 letter spaces at 4 letter positions from fixation. We conclude that local signs for letters become increasingly imprecise in peripheral vision. This imprecision may account for errors in judging the order of letters in text, which may in turn, have an impact on reading.