This paper presents a watermarking algorithm that is suitable for the data represented in floating- or fixed-point numbers. By keeping the modulation information, every value in the original object can be restored with the smallest error. For the sake of exact recovery, the redundancy in the watermarked object should be exploited to save the recovery information. We analyze the security of the algorithm and enhance it by adding a pseudorandom (PR) sequence to the watermarked object. The algorithm can be applied in the measurement to embed some useful information in a distortion-free way. We implement it on the Virtual Reality Modeling Language (VRML) models, and the experimental results show its efficacy.