Aim: Age determination is one of the most common forensic procedures which has a wide range of applications, including identifications of accident victims, crime investigations, and social benefit regulations. To date, only a few studies have employed Raman spectrometry to relate the aging-dependent compositional changes in tooth tissues with the donor’s age for forensic purposes. The available studies are limited to investigations of intact teeth, representing an ideal-case scenario which is quite different from forensic reality. Therefore, the aim of this study was to simulate a more realistic forensic scenario and evaluate the feasibility of Raman spectrometric age determination by using teeth affected by various pathologies. Also, tooth specimens were used for the collection of Raman spectra without any previous preparation, as non-destructive specimen handling is generally preferred in forensic investigations. Materials and methods: The sample of 71 teeth used for this study was obtained by a random draw from the archive of the Department of Dental Anthropology of the School of Dental Medicine, University of Zagreb, Croatia. The age of tooth donors ranged between 11 and 76 years. The teeth had been extracted due to various indications, most common being periodontitis (51 %) and failed endodontic treatment (39 %). To simulate a forensic analysis of teeth at different post-extraction time periods, the time span between extraction and performing Raman spectrometric measurements ranged between 0.1 and 5.5 years. No special selection criteria were applied; teeth affected with various pathological processes were deliberately included to simulate a realistic sample. Raman spectra were recorded using an FT-Raman accessory of the Spectrum GXspectrometer (Perkin-Elmer, Waltham, MA, USA) equipped with an Nd-YAG laser of 1064 nm wavelength. Each spectrum was recorded by averaging 100 scans in the spectral range between 3500 and 200 cm-1 and with a spectral resolution of 4 cm-1 . Spectra were collected from three distinct sites on each tooth: crown, neck, and apex. The spectra were stored in a dataset and connected with the donor’s age and collection site. All spectra were baseline corrected and normalized using the peak at 960 cm-1 (symmetric PO4 stretching) to exclude possible differences caused by variations in recording conditions. Whole Raman spectra (3500 – 200 cm-1 ) were used for principal component analysis (PCA) and principal component regression (PCR) with 3 to 7 principal components. PCR was used to establish a relationship between the recorded spectra and the donor’s age. PCA and PCR were performed using a model built within Matlab 2010 (MathWorks, Natick, MA, USA) and its add-on PLS_Toolbox (Eigenvector Research, Manson, WA, USA). Separate PCA/PCR models were built according to the spectra collection site and donor’s gender, resulting in the following six combinations: apex male, apex female, crown male, crown female, neck male, and neck female. Additionally, spectra from both genders were mixed together in three “joint” models that accounted only for the spectra collection site. All models were cross-validated using the Venetian blinds method. Results: Raman spectra of dental hard tissues featured the inorganic part represented by vibrational bands in the wavenumber range of 1100 – 400 cm-1 (PO4 and CO3 vibrations) and the organic part represented in the range of 3100 – 1100 cm-1 (amide bands and C-H vibrations). The inorganic/organic ratio was considerably higher in spectra collected from the crown compared to the other two sites. The changes in the intensity of bands representing both the inorganic and organic components of tooth tissues indicated that it is possible to build a PCA model for distinguishing donor’s age using Raman spectra. Also, the spectral changes occurring as a function of donor’s age indicates that a PCR model for age determination can be built. The predictive capabilities of PCR models for age determination differed according to different spectra collection sites. The highest coefficients of correlation and lowest error values were obtained in models based on apex spectra (R2 values of 0.84 and 0.71 for males and females, respectively). Models based on other combinations of genders and spectra collection sites had comparably lower R2 values, ranging from 0.17 to 0.59. Low R2 values (0.18 – 0.24) were also obtained for joint models that were built using spectra from both genders. The PCA models based on apex spectra showed a comparatively better separation of age groups than the models based on neck and crown spectra. Accordingly, the PCR age determination models based on apex spectra showed higher coefficients of determination and lower error values compared to the models based on neck and crown spectra. Conclusions: This study showed that the age determination model based on principal component regression can be built using Raman spectra collected from surfaces of non-sound teeth without any previous preparation. The optimal age determination capability was attained by using Raman spectra collected from cementum at root apex, whereas spectra collected from mineralized tissues at the tooth neck and crown were less suitable. The age determination model based on apex spectra showed an optimal performance only when tooth donor genders were analyzed separately.