Freight demand can be modelled either as Freight Trip Generation (FTG) or Freight Generation (FG). While in FG modelling the modelling unit is tonnage or pallets, in FTG models it is vehicle trips. Data employed in freight demand models are predominantly obtained through Establishment-Based Freight Surveys (EBFS). Despite its widespread use, EBFS suffer from unit and item nonresponse. While unit nonresponse is well studied, item nonresponse has received little attention. Majority of the FTG modelling studies employed Simple Linear Regression (SLR) which does not capture all the FTG characteristics. Count models that consider more characteristics than Multiple Linear Regression (MLR) models are rarely employed. Besides, majority of the studies segment FTG by establishment type. They consider only trips by trucks or assume that all trips are equivalent to a truck trip. However, freight in several large cities is increasingly being moved by smaller vehicles, requiring models segmented by vehicle type. FTG models also rarely consider spatial dependence. Lastly, urban FG modelling studies are rare and have focussed on large urban freight traffic generators. They ignored establishments typical to central urban environments. The present study aims to fill these gaps. The study analyses the item nonresponse in EBFS and proposes oversampling a few establishment groups to handle item nonresponse. The proposed technique is compared with pair-wise deletion, single imputation by mean, median, and mean imputation by group. The study calculates FG and FTG rates per area and employee. Further, it develops two sets of models for FTG segmented by vehicle type, one without and the other considering spatial dependence. The spatial Zero-Inflated Negative Binomial (ZINB) model, which captures most FTG characteristics, provides the best fit for FTG in most cases. The study also quantified the bias in model estimates when segmentation by vehicle type, spatial dependence, error-term correlations, and FTG characteristics are ignored using elasticities The study also develops SLR and Proportional Odds logit models for FG using area, employment and operational age as predictors. The area model is the best FG model.