This study examines the influence of socio-demographic and physical features on residential water demand in Bojnourd, a semi-arid city of 228,931 residents in northeastern Iran. Using billing records from the North Khorasan Water and Wastewater Company, GIS‑based physical data from the Bojnourd Municipality, and census data from the Statistical Center of Iran, we compiled two administrative datasets after excluding implausible entries: 1,660 census blocks (2011; socio‑demographic predictors: population density, age, education) and 28,218 residential parcels (2016; physical predictors: building size, building age, number of stories). Each dataset was split into training (70%) and test (30%) subsets. Correlation analysis, ordinary least squares (OLS) regression, and spline regression were performed in SPSS and R, followed by logistic regression to model thresholds of high consumption (>180 m³/year). At the parcel level (2016), the number of stories showed a moderate negative correlation with water consumption (r = -0.35, p < 0.001) and, along with building size, emerged as a significant predictor of high consumption in the logistic regression model (OR = 0.279 and 1.012, respectively); building size and building age showed statistically significant but practically negligible bivariate correlations with consumption (r = 0.03). At the census-block level, population density was negatively correlated with per-capita consumption (r = -0.31), while education (r = 0.50) and age (r = 0.53) were positively correlated; spline regression explained 41.81% of variance (R² = 0.4181) at this level, compared with 23.62% (R² = 0.2362) at the parcel level, and revealed diminishing returns of education on consumption at higher attainment levels. As these two analyses draw on data from different years and levels of aggregation, they are presented as complementary but independent lines of evidence rather than a single integrated model. These findings offer strategies for sustainable urban water management in semi-arid cities of less developed countries.