Quantifying the full 3-D shape of the human crystalline lens is important for improving intraocular lens power or sizing calculations in treatments of cataract and presbyopia. In a previous work we described a novel method for the representation of the full shape of the ex vivo crystalline lens called eigenlenses, which proved more compact and accurate than compared state-of-the art methods of crystalline lens shape quantification. Here we demonstrate the use of eigenlenses to estimate the full shape of the crystalline lens in vivo from optical coherence tomography images, where only the information visible through the pupil is available. We compare the performance of eigenlenses with previous methods of full crystalline lens shape estimation, and demonstrate an improvement in repeatability, robustness and use of computational resources. We found that eigenlenses can be used to describe efficiently the crystalline lens full shape changes with accommodation and refractive error.