# How long are screenplay scenes?

**Edition 1.0 · Published 4 November 2026 · Script Intelligence · Original quantitative study**

Across 344 screened feature screenplays, the median of each script’s median scene length is **half a modelled page**. The middle half lies between **0.375 and 0.625 modelled pages**. These findings describe the available scripts in this sample; they do not set an ideal scene length.

We measured the typical scene within each script, then compared those script medians with equal weight. A script with 300 scenes therefore has the same weight as a script with 50. The result is not the median of every scene pooled together.

## What a modelled page means

The engine uses fixed text widths and spacing, with 55 lines per page. Scene headings are included. Each detected scene is converted to eighths of a page, rounded to the nearest eighth with a minimum of one eighth. It then calculates the script’s median and mean. The resulting lengths are neither physical spans in the original PDF nor estimates of screen time.

A detected scene follows the parser’s recognised headings. Changes in draft formatting, heading conventions or extraction quality can affect where the parser divides scenes. Source screening reduces known problems but does not establish that every boundary is correct.

## The distribution

| Measure across script medians | Modelled pages |
|---|---:|
| Median | 0.5 |
| Middle half | 0.375–0.625 |
| 10th to 90th percentile | 0.25–0.875 |
| Observed range | 0.125–3.75 |

316 of 344 scripts, **91.9%**, have a script median below one modelled page. That is a statement about script medians, not the percentage of individual scenes shorter than a page.

Interactive charts: https://www.theprelist.io/research/scene-length/versions/scene-length-2026-11-v1

Frozen figures: [Distribution](figures/scene-length-distribution.png) · [Genre comparisons](figures/scene-length-genres.png). The genre bands show the middle half, with the median marked; all values are in modelled pages.

## Mean and median answer different questions

Within every eligible script, the mean scene length is above its median. The median within-script difference is approximately **0.39 modelled pages**. Longer scenes pull the mean upwards, so it is useful to distinguish the average scene length from the middle scene length when comparing drafts.

This does not establish how scene length affects audience attention, writing quality, production cost or commercial success. Those questions require different evidence.

## Genre comparisons

These are descriptive primary-genre groupings. Genres overlap, and one editorial label does not capture everything a film does. The minimum of 20 scripts is a reporting rule, not proof of representativeness or statistical significance.

| Primary genre | Eligible scripts | Median script median, modelled pages |
|---|---:|---:|
| Action | 36 | 0.25 |
| Animation/Family | 33 | 0.625 |
| Comedy | 56 | 0.5625 |
| Crime | 12 | Withheld |
| Drama | 87 | 0.375 |
| Fantasy | 14 | Withheld |
| Horror | 39 | 0.5 |
| Romance | 20 | 0.5 |
| Sci-Fi | 26 | 0.375 |
| Thriller | 21 | 0.375 |

## Source screening and sensitivity

The same 428 v5 source PDFs were remeasured with extraction method 1.6. One incomplete source, *Saw*, was excluded entirely. Scene measurements are withheld for prior scene-family holds, unresolved image content and discrepancies with an independent heading-candidate count. Seventy-five flagged front-matter pages in 69 files were visually classified; clearing eligible front-matter flags did not clear unrelated measurement concerns.

The median and middle-half range remain 0.5 and 0.375–0.625 in the high-confidence-only subset (n=335) and the subset without dialogue-family flags (n=331). Fifteen included script medians change under method 1.6 compared with 1.5. A stable headline does not mean every input was unchanged.

## Scope and limits

The input is the same 428 available feature-script PDFs used in the September v5 release. After excluding the incomplete source, 427 remain in the corrected benchmark. This study uses 344 with eligible scene measurements. The 84 exclusions from the original input are listed in [the review queue](analysis/review-queue.csv); reasons can overlap.

The 344 scripts cover released features from 2000 to 2026. Public availability and awards circulation influence which drafts survive in the library. This is not a random sample of produced films or submitted screenplays. Draft versions, editorial genre labels and extraction errors remain sources of uncertainty. TV is outside this study.

Independent numerical checks reproduce all **46,691 per-scene numeric records** used here, including rounding, per-script means and medians, overall quantiles and eligible genre results. These checks establish calculation consistency. The earlier 72 selected boundary matches and the new front-matter review are limited source checks, not a whole-corpus accuracy rate.

## Reproduce and version

Run `python3 analysis/analyse.py`, then `python3 review/verify-numerics.py` with NumPy installed. Inputs are pinned by SHA-256 in `inputs/manifest.json`; [publication.json](publication.json) records the edition, source release and protocol. Source texts are not included in this study package.

This first published edition is frozen at https://www.theprelist.io/research/scene-length/versions/scene-length-2026-11-v1. Review is due by 4 February 2027, or sooner after a material source or method change. Corrections and expanded samples receive new editions with before/after notes; existing files and citations remain available. Reviews do not imply automatic data updates.

The derived data and article are licensed CC BY 4.0. Screenplay text and third-party imagery are excluded. This is an internal research review, not external academic peer review.