We present and characterize the galaxy-galaxy lensing signal measured using the first three years of data from the Dark Energy Survey (DES Y3) covering 4132 deg2 . These galaxy-galaxy measurements are used in the DES Y3 3 ×2 pt cosmological analysis, which combines weak lensing and galaxy clustering information. We use two lens samples: a magnitude-limited sample and the redMaGiC sample, which span the redshift range ∼0.2 - 1 with 10.7 and 2.6 M galaxies, respectively. For the source catalog, we use the METACALIBRATION shape sample, consisting of ≃100 M galaxies separated into four tomographic bins. Our galaxy-galaxy lensing estimator is the mean tangential shear, for which we obtain a total SNR of ∼148 for MagLim (∼120 for redMaGiC), and ∼67 (∼55 ) after applying the scale cuts of 6 Mpc /h . Thus we reach percent-level statistical precision, which requires that our modeling and systematic-error control be of comparable accuracy. The tangential shear model used in the 3 ×2 pt cosmological analysis includes lens magnification, a five-parameter intrinsic alignment model, marginalization over a point mass to remove information from small scales and a linear galaxy bias model validated with higher-order terms. We explore the impact of these choices on the tangential shear observable and study the significance of effects not included in our model, such as reduced shear, source magnification, and source clustering. We also test the robustness of our measurements to various observational and systematics effects, such as the impact of observing conditions, lens-source clustering, random-point subtraction, scale-dependent METACALIBRATION responses, point spread function residuals, and B modes.