This paper proposes a distributed multiscale data compress algorithm which can transform irregular sample data. Considering the characteristics and location information of nodes in sensor networks, a noveldistributed domain partition mode DDPM (distributed domain partition model) is proposed first. On the basis of thismodel, a multiscale data compress model—MDCM (multiscale data compress model) is proposed for sensornetworks. MDCM uses Voronoi tessellation partition the domain created by DDPM. Theoretical analyses and simulation results show that the novel methods above have good ability of approximation, and can compress the data efficiently, reduce the amount of data greatly.