Structural sensitivities are often required in dynamic analyses of engineering structures, such as system identification and control, structural modification and optimization. Many different methods have been developed in the last three decades for the efficient computation of such sensitivities. Though these existing methods have proven to be very useful tools to structural analysts, they are restricted to those cases where accurate analytical or finite element models are available. In many practical applications where sensitivities are needed in the solution of troubleshooting problems only limited measured data are available; existing methods being inapplicable. In this paper, a new and effective method has been developed to derive structural design sensitivities, both frequency response function sensitivities and eigenvalue and eigenvector sensitivities, from limited vibration test data. Design sensitivities calculated directly from measured data are more accurate than those calculated from analytical or finite element models since structural modelling errors are inevitable due to the complexity of almost all engineering structures. The relationship between frequency response function sensitivities and eigenvalue and eigenvector sensitivities has been established. This relationship provides the theoretical basis for the experimental determination of eigenvalue and eigenvector sensitivities. The effect of residue (contribution of out-of-range modes) upon analysis accuracy has been examined. Detailed numerical assessment based on a turbine bladed disk model as well as experimental investigation of a beam structure have been conducted and the results have demonstrated the practicality of the proposed method.