India has gradually progressed into fertility transition over the last few decades. However, the timing and pace of this transition has varied notably in terms of both its geography and the demographic groups most affected by it. While much literature exists on the relationships between fertility level and its influence on demographic, economic, socio-cultural and policy-related factors, the potential spatial variations in the effects of these factors on the fertility level remain unaddressed. Using the most recent district-level census data (of 2011) for India, this nationwide study has identified plausible spatial dependencies and heterogeneities in the relationships between the district-wise Total Fertility Rates (TFRs) and their respective demographic, socioeconomic and cultural factors. After developing a geocoded database for 621 districts of India, spatial regression and Geographically Weighted Regression (GWR) models were used to decipher location-based relationships between the district-level TFR and its driving forces. The results revealed that the relationships between the district-level TFR and the considered selected predictors (percentage of Muslims, urbanization, caste group, female mean age at marriage, female education, females in the labour force, net migration, sex ratio at birth and exposure to mass media) were not spatially invariant in terms of their respective strength, magnitude and direction, and furthermore, these relationships were conspicuously place- and context-specific. This study suggests that such locality-based variations and their complexities cannot be explained simply by a single narrative of either socioeconomic advancement or government policy interventions. It therefore contributes to the ongoing debate on fertility research in India by highlighting the spatial dependence and heterogeneity of the impacts made by demographic, socioeconomic and cultural factors on local fertility levels. From a methodological perspective, the study also discerns that the GWR local model performs better, in terms of both model performance and prediction accuracy, compared with the conventional global model estimates.