SPAD combined with hyperspectral sensors can be an alternative approach to traditional laboratory methods for determining the physiological status of trees. The aim of this work was to assess whether the effectiveness of SPAD predictive models using hyperspectral data might be influenced by where the measurements were carried out. Leaves of apricot trees of two varieties (Farbaly and Farlis) were analysed with SPAD and spectroradiometer, and the data were organized in two different ways: (i) overall dataset (OD), collecting total measurements of trees in each variety; (ii) subset of overall datasets (SOD), collecting the measurements performed on the cardinal points of trees in each variety. Prediction models were built using as regressors: (i) spectral data transformed with Continuum Removal (CR) methodology (CR indices); (ii) vegetation indices (VI) linked to chlorophyll and nitrogen content; (iii) reflectance values associated with chlorophyll content and to wavelengths ranges where (CR) methodology was applied; (iv) reflectance values of full spectrum. The best performances belonged to models using wider ranges of spectrum both in ODs and in SODs. The north cardinal point showed prediction models with the best performances in both varieties. No VI and CR indices showed reliable models. All the reliable prediction models were associated with compounds involved in physiological state and metabolism of leaves in apricot tree.