A data recovery apparatus and method by using over-sampling are provided. The data recovery apparatus by using over-sampling includes an over-sampling module, a data regeneration unit, a phase alignment unit, a phase decision module, and an output data correction unit. The over-sampling module samples serial data according to a clock signal, so as to output M-bit data, in which each bit in the serial data is sampled N times. The phase alignment unit selects specific M-bit data from a P-bit signal output by the data regeneration unit, and distinguishes the specific M-bit data to X groups of N-bit signals. The phase decision module determines a direction of phase adjustment according to the specific M-bit data. The output data correction unit selects and outputs first or second recovery data constituted by first or second specific bits of each group of N-bit signal according to the direction of phase adjustment.