<p>This paper presents a novel method for synthesizing whole-body humanoid motions by blending different basic components under physical constraints based on Functional Principal Component Analysis (FPCA). After a process called smoothing that approximates measured human motions by B-Spline functions satisfying physical constraints, we apply FPCA to extract a small number of principal motion components that can be combined to generate different motions. The proposed method allows visualizing whole-body motions of many degrees of freedom (DOFs) in a compact FPC space, and synthesizing new physically consistent motions by applying physical constraints to those motions blended from principal components.</p>