Structural design of float glass is performed with the semi-probabilistic approach by defining partial factors for glass strength, calibrated with full probabilistic methods on paradigmatic case-studies. Although the most used statistics for float-glass strength is the 2-parameter Weibull’s (2PW), we show that it fails to interpret the extreme tail of the population of experimental data, associated with small failure probabilities. The tails govern the calibration of partial factors, since the target values for failure probability admitted for a construction work are very small; hence, this statistics leads to very conservative design. Contrariwise, from experiments a lower bound for float-glass strength, attributable to modern factory production controls, is recognizable. Moreover, through a micro-mechanically-motivated interpretation, we show that corrosion/abrasion do not imply the decay of strength below a certain limit, inducing instead a lower dispersion in the data. Therefore, we propose to use generalized Weibull statistics, among which the left-truncated Weibull distribution, to better interpret the tail of the population. Partial material factors calibrated with the novel approach are considerably lower than with the 2PW statistics. Enormous saving could be achieved with this more refined statistical approach to interpret float-glass strength.