The relationship between biodiversity and ecosystem multifunctionality (EMF) is crucial for understanding the processes of ecological restoration in semi-arid regions. However, partitioning the relative influence of various biodiversity attributes, namely taxonomic, functional, and phylogenetic diversity, on EMF during secondary succession is still unclear. This study aimed to bridge the gap by employing field measurements and the chronosequence approach at 21 plots with different stand ages and precipitation conditions on the Loess Plateau of China. For diversity indices, we calculated the Shannon–Wiener diversity index, Simpson’s dominance index, Pielou evenness index, community weighted mean (CWM), functional variance (FDvar), and Faith’s phylogenetic diversity (PD) based on the empirically measured composition and traits of plant species. The EMF was expressed as the averaged value of eight function variables (including aboveground biomass, root biomass, soil total carbon, total nitrogen, and total phosphorus content, soil organic carbon, available nitrogen and available phosphorus content). The results showed that species evenness and CWM of leaf dry matter content (LDMC) significantly increased yet the CWM of specific leaf area (SLA) decreased with stand age, indicating the resource-use strategy of the plants became more conservative through succession into its later stages. The EMF increased with both stand age and mean annual precipitation. The structural equation model revealed that stand age, soil water content (SWC), and the multiple diversity indices altogether accounted for 56.0% of the variation in the EMF. PD and the CWMs of plant height and LDMC had positive effects on the EMF, and the FDvar of leaf nitrogen had negative effects on EMF. However, the Shannon Wiener diversity had no significant effect on the EMF. Our results suggest that functional and phylogenetic diversity are more important than taxonomic diversity in predicting EMF, and that multidimensional biodiversity indices should be jointly considered to better predict EMF during the succession of semiarid grasslands.