The structural remodeling of collagen is important in several biological processes such as wound healing, tendon repair, fibrosis and developmental morphogenesis. Multiphoton microscopy, which uses ultra-short femto-second laser pulses as an excitation source, is efficient in the multiphoton excitation fluorescence (MPEF) of exogenous fluorescent labels tagged to various cellular macromolecular objects, as well as in the induction of a highly specific second harmonic generation (SHG) signal from non-centrosymmetric macromolecules such as fibrillar collagens. Although the non-descanned detectors in the reflection geometry have normally been employed for capturing the backward scattered SHG as well as the MPEF signals, considering the wide range of engineered thick tissue imaging applications, there are still un-answered questions about the generated 3D collagen structures because of the directional pattern of SHG signals. The present study dealt with an in vitro collagen-fibroblast raft model in which the stimulation of fibroblast cells induced the lateral orientation of collagen molecules. The SHG signals originating from the 3D collagen matrix were captured simultaneously in both forward and backward scattering directions to understand the collagen structural differences and to generate a comprehensive understanding of collagen matrix remodeling. The cells were stained for cellular actin with phalloidin conjugated to Alexa Fluor 488. Non-descanned detectors in both reflection and transmission geometries, as well as the spectral scanning mode in the reflection geometry, were employed for generating the 3D images and SHG spectral information, respectively. We found that spectrally clean SHG signal peaked at 425-nm with an excitation wavelength of 850-nm and the same excitation wavelength that was used for Alexa 488 excitation. For the collagen raft thickness investigated, we have shown here that both forward and backward scattered SHG signals decayed in a similar way as we move deep into the collagen raft, despite the fact that the 3D SHG image collected in the backward direction was relatively diffusive compared to that collected in the forward direction. The measured line and colocalization profiles, the colocalization coefficients and the 3D voxel volumes altogether indicated a higher degree of correlation of both the forward and backward SHG signals. In conclusion, our analysis of forward and backward 3D SHG image data sets generated from 3D collagen raft models indicates the backward SHG detection, which is practical for thick engineered tissue sections, provides all of the relevant quantitative details as those of the forward scattered SHG.