Intelligence is often defined as the ability of an agent to learn, adapt to its environment, and solve novel challenges. However, despite over 100 years of theoretical development (e.g., general intelligence), widespread explanatory power (up to 50% of variance in cognitive scores), and the ability of intelligence measures to predict important life outcomes such as educational achievement and mortality, the exact configuration and neural correlates of cognitive ability remain poorly understood. This dissertation aims to make progress in this pursuit by exploring how human brain structure and intelligence correlate and co-develop with each other from childhood to early adulthood (ages 5 - 22 years). This endeavour is undertaken in three large cohorts (N range: 337 - 2072), guided by theory (e.g., crystallised and fluid intelligence), and implemented using rigorous, cutting-edge quantitative methods (i.e., structural equation modelling and network science). The results of this research provide robust evidence that the brain-behaviour relationships in intelligence are complex (i.e., consists of many independent yet interacting parts) and change nonlinearly during development. The first study sought to elucidate the factorial structure and white matter substrates of child and adolescent intelligence using two cross-sectional, developmental samples (CALM: N = 551 (N = 165 neuroimaging), age range: 5 - 18 years; NKI-Rockland: N = 337 (N = 65 neuroimaging), age range: 6 - 18 years). In both samples, it was found (using structural equation modelling (SEM)) that cognitive ability is best modelled as two separable yet related constructs, crystallised and fluid intelligence, which became more distinct (i.e., less correlated) across development, in line with the age differentiation hypothesis. Further analyses revealed that white matter microstructure, most prominently of the superior longitudinal fasciculus, was strongly associated with crystallised (gc) and fluid (gf) abilities. Finally, SEM trees, which combines traditional SEM with decision trees, provided evidence for developmental reorganisation of gc and gf and their white matter substrates such that the relationships among these factors dropped between ages 7 - 8 years before increasing around age 10. Together, these results suggested that shortly before puberty marks a pivotal phase of change in the neurocognitive architecture of intelligence. The second study builds upon the first by again examining the neurocognitive structure of intelligence, this time from a network perspective. The network or mutualism theory of intelligence presupposes direct (statistical) interactions among cognitive abilities (e.g., maths, memory, and vocabulary) throughout development. Therefore, this project used network analytic methods (specifically graphical LASSO) to simultaneously model brain-behaviour relationships essential for general intelligence in a large (behavioural, N = 805; cortical volume, N = 246; fractional anisotropy, N = 165), developmental (ages 5 - 18 years) cohort of struggling learners (CALM). Results indicated that both the single-layer (cognitive or neural nodes) and multilayer (combined cognitive and neural variables) networks consisted of mostly positive, small partial correlations, providing further support for the mutualism/network theory of cognitive ability. Moreover, using community detection (i.e., the Walktrap algorithm) and calculating node centrality (absolute strength and bridge strength), convergent evidence suggested that subsets of both cognitive and neural nodes play an intermediary role 'between' brain and behaviour. Overall, these findings suggest specific behavioural and neural variables that may have greater influence among (or might be more influenced by) other nodes within general intelligence. The final study investigated the longitudinal relationships between human cortical grey matter structure and measures of decision-making, risk-related behaviours, and spatial working memory from adolescence to early adulthood (ages 14 - 22 years). In the IMAGEN study (maximum N across time points/waves = 2072), latent growth curve models were used to estimate the baseline and longitudinal associations between behavioural measures and cortical surface area, thickness, and volume. Univariate models (only behavioural or neural measures) revealed that performance in decision-making, risk-related behaviours, and spatial working memory, as well as brain structure changed nonlinearly from mid-adolescence (age 14) to early adulthood (age 22). Furthermore, bivariate models (combined behavioural and neural measures) provided evidence for adaptive reorganisation (behaviour intercept predicts changes in brain structure) but not structural scaffolding (brain structure intercept predicts changes in behaviour). Furthermore, findings suggested that there were no correlated changes between behavioural and brain structure slopes (rates of change from mid-adolescence to early adulthood). This dissertation concludes by summarising the core results, addressing key limitations, and discussing avenues for future research. Taken together, this thesis hopes to convince cognitive neuroscientists that to understand cognitive ability and its neural determinants, they (we) must work more diligently toward building coherent, rigorous, and testable neurocognitive theories of intelligence-particularly under the conceptual and analytic paradigm of complex systems.