Methods and apparatus are provided for determining a position of an offset latch employed for decision-feedback equalization. The position of an offset latch is determined by obtaining a plurality of samples of a data eye associated with a signal, the data eye comprised of a plurality of trajectories for transitions out of a given binary state; determining an amplitude of at least two of the trajectories based on the samples; and determining a position of an offset latch based on the determined amplitudes. The initial position of the offset latch can be placed, for example, approximately in the middle of the determined amplitudes for at least two of the trajectories. The initial position of the offset latch can be optionally skewed by a predefined amount to improve the noise margin.