This paper is a sum and some development of ideas studied in previous author papers devoted to investigation of fiber scale effect (strength-length function). In first part of the paper some theoretical aspects of the problem and in the second an application to the processing of the test dataset are discussed. As distinct from our previous publications, two types of defects (“technological”, i. e. independent on load, and load-dependent), two types of the fiber strength dependences on defect number are introduced and the probability of absence of defects is taken into account. We consider specimen as sequence of n elements with the same length. It is supposed that there are defects in K of them, 0≤K≤ n. Two cases are considered: number of defects, K, is a random variable or a random process, . In last case the process of increasing of K and failure of specimens is described as Markov chain, matrix of transition probabilities of which depends on the current value of the process of loading described as some (ascending up to infinity) sequence { , ,…, , …}. Three versions of the specimen strength dependence on the number of defected elements are considered for both cases. So six probability structures are introduced and different versions of specified distribution functions and corresponding models are considered. The methods of the model parameter estimation, results of processing of glass, flax, carbon fiber and carbon bundle datasets, comparison of the different models are presented in second part of the paper.