Information signals such as grayscale images or audio signals are represented as a sequence of PCM signal samples. To embed auxiliary data in the least significant bits of the signal, the samples are slightly distorted. There is a so-termed "rate-distortion function" (20) which gives the largest embedding rate R given a certain distortion level D. It appears that the efficiency of prior art embedding schemes such as LSB replacement (21,22) can be improved. The invention discloses such embedding schemes (23,24). According to the invention, the signal is divided into groups of L (L>1) signal samples (x). For each group of signal samples, a vector of least significant portions (x mod n) of the signal samples is created. For n=2, the vector comprises the least significant bit of each signal sample. The syndrome of said vector (as defined in the field of error detection and correction) represents the embedded data. Only one (or a few, in any case less than L) signal sample(s) of a group needs to be modified so as to achieve that the vector assumes a desired syndrome value.