Across the fixed 23-country OECD comparison set, reading and mathematics scores were more than 30 points lower in 2025 than at the start of their comparable PISA series. Science also declined, but its long-run drop was smaller. The matched-country series matters because the membership used for the average stays constant across assessment cycles.
The long-run OECD trend turned lower after the early 2010s
Reading is comparable from 2000, mathematics from 2003, and science from 2006. The values below use the OECD average-23, the fixed set used by the OECD for comparisons spanning the full PISA history.
Reading, mathematics and science scores across PISA cycles
OECD average-23, from each subject’s first comparable cycle through PISA 2025
OECD PISA 2025 Database, Tables I.B1.2a.36–38. The vertical scale starts at 440 so cycle movement is readable; PISA mean scores are not ratio-scale measures.
Reading reached its matched-series high in 2012 and then fell in every later cycle. Mathematics stayed within a relatively narrow band through 2018 before dropping sharply. Science peaked in 2009 and declined more gradually; its 2022-to-2025 movement was not statistically distinguishable from zero in the OECD comparison.
OECD average-23 mean scores by PISA cycle
The dashes mark years before a subject’s long-run comparable scale begins.
Scroll horizontally to view all columns.
| PISA cycle | Reading | Mathematics | Science |
|---|---|---|---|
| 2000 | 499.5 | — | — |
| 2003 | 497.4 | 502.1 | — |
| 2006 | 495.5 | 501.1 | 502.7 |
| 2009 | 499.4 | 501.7 | 505.7 |
| 2012 | 501.4 | 498.7 | 504.7 |
| 2015 | 497.2 | 495.5 | 497.0 |
| 2018 | 492.8 | 496.4 | 493.5 |
| 2022 | 481.9 | 480.3 | 490.9 |
| 2025 | 465.8 | 469.4 | 485.9 |
Values are OECD average-23 estimates and are shown to one decimal place. Differences calculated from rounded values can vary slightly from changes calculated by the OECD from unrounded estimates.
The post-2018 break is clearest in reading and mathematics
The same long-run series shows that much of the decline in reading and mathematics is concentrated in the two cycles after 2018.
| Subject | Long-run start | Start to 2025 | 2018 to 2025 | 2022 to 2025 |
|---|---|---|---|---|
| Reading | 2000 | −33.7 | −27.0 | −16.1 |
| Mathematics | 2003 | −32.7 | −27.0 | −10.9 |
| Science | 2006 | −16.8 | −7.5 | −4.9 |
Score-point changes use the OECD trend tables. The science change from 2022 to 2025 is not statistically distinguishable from zero.
The timing differs by subject
Reading had already moved below its 2012 high before 2018, but the later decline became much steeper. Mathematics shows a sharper break in the series: the matched OECD average was 496.4 in 2018, only 5.7 points below 2003, then fell another 27 points by 2025. Science moved downward too, but not at the same pace.
Lower-achieving students lost more ground over the last decade
Mean scores do not show where within the performance distribution the movement occurred. OECD percentile data add a second view of the 2015–2025 decline.
Score decline at the 25th and 75th percentiles, 2015–2025
OECD average-23; bars show the magnitude of the ten-year decline in score points
OECD PISA 2025 Database, Tables I.B1.2a.39–41. Larger bars represent larger score declines between 2015 and 2025.
The 25th percentile fell 7 points more than the 75th percentile in science, 11 points more in reading and 9 points more in mathematics. Both ends moved lower, but the steeper decline among lower-achieving students widened the distance between the two parts of the distribution.
Why the 2025 trend values differ from the headline OECD averages
Two valid OECD averages answer different questions. The long-run series holds the comparison group constant; the headline 2025 average describes the current OECD membership included in the 2025 comparison.
OECD membership and data availability changed over time. Comparing a current-cycle OECD average directly with an older average built from a different country set can mix performance change with membership change. The OECD average-23 avoids that problem for a 2000–2025 trend.
Data basis and trend comparability
OECD PISA 2025 Annex B1 contains the long-run mean-score tables used here: Table I.B1.2a.36 for science, I.B1.2a.37 for reading and I.B1.2a.38 for mathematics.
PISA 2025 Results, student performance chapter reports the 25-year trajectory, including the reading peak around 2012, the science peak in 2009 and the sharper mathematics decline after 2018.
PISA 2025 Reader’s Guide defines OECD average-23 as the fixed 23-member arithmetic average that supports comparisons across all PISA cycles from 2000 to 2025. It also explains rounding and changing country coverage in other OECD averages.
PISA 2025 multidimensional student learning chapter reports the 2015–2025 changes at the 25th and 75th percentiles used in the distribution chart.
OECD PISA Student Scores dashboard reports the 2025 headline OECD averages, while the Performance Trends dashboard provides an interactive view of changes over time.
PISA is designed as a three-year cycle, but the OECD PISA program page records two pandemic-era postponements: the planned 2021 assessment moved to 2022, and the planned 2024 assessment moved to 2025. The trend therefore uses the actual assessment years rather than assuming equal three-year spacing.